Temperature, Pressure, and Chemical Potential as Entropy Derivatives

Here is a question that sounds almost too basic to ask: what is temperature? Not how you measure it — thermometers and mercury and expansion — but what it fundamentally is. The surprising and beautiful answer of statistical thermodynamics is that temperature is a derivative of entropy: $1/T = (\partial S/\partial U)_{V,N}$ , the rate at which a system's entropy rises as you feed it energy. Pressure and chemical potential are two more entropy slopes. All three intensive variables — the quantities that equalize when systems reach equilibrium — are simply the partial derivatives of the single fundamental relation $S(U,V,N)$ . Why should temperature be $\partial S/\partial U$ rather than something else? Because that is what the second law forces. Bring two isolated systems into thermal contact and let them share energy at fixed total energy. The total entropy is maximized at equilibrium, and maximizing $S_1(U_1) + S_2(U_2)$ under $U_1 + U_2 = \text{const}$ requires $(\partial S_1/\partial U_1) = (\partial S_2/\partial U_2)$ . Some derivative of entropy is equal across the two systems at equilibrium — and that is exactly what we mean by "same temperature" (the zeroth law). Identifying the common value with $1/T$ pins down the definition. The same argument with volume exchange defines $P/T$ , and with particle exchange defines $-\mu/T$ . This is not merely a redefinition; it explains the direction of natural processes. Because entropy is a concave function of energy — its second derivative is negative — the entropy-maximum is genuinely a peak, and the approach to it forces energy to flow from the body with the smaller $\partial S/\partial U$ (higher $T$ ) to the one with the larger (lower $T$ ). Heat flows from hot to cold not by decree but because that is the direction that raises total entropy. The sign of a second derivative encodes the arrow of thermal processes. Assembling the three slopes gives the entropy-representation fundamental relation $\mathrm{d}S = \tfrac1T\,\mathrm{d}U + \tfrac PT\,\mathrm{d}V - \tfrac\mu T\,\mathrm{d}N$ , which rearranges into the familiar $\mathrm{d}U = T\,\mathrm{d}S - P\,\mathrm{d}V + \mu\,\mathrm{d}N$ . These derivatives are the gateway from the abstract entropy surface to every measurable thermodynamic quantity, and they are where the statistical definition of temperature (via $S = k_B\ln\Omega$ ) meets the thermodynamic one.

Key takeaways

The three intensive variables are entropy slopes: $\tfrac1T = (\partial S/\partial U)_{V,N}$ , $\tfrac PT = (\partial S/\partial V)_{U,N}$ , $-\tfrac\mu T = (\partial S/\partial N)_{U,V}$ .
These definitions are forced by the second law: maximizing total entropy under exchange of a conserved quantity equalizes the corresponding entropy derivative across subsystems.
Equal $\partial S/\partial U$ ⟺ equal temperature (zeroth law); equal $\partial S/\partial V$ at fixed $T$ ⟺ equal pressure; equal $\partial S/\partial N$ ⟺ equal chemical potential.
Entropy is concave in energy ( $\partial^2 S/\partial U^2 < 0$ ), so the extremum is a genuine maximum and heat flows hot → cold — the sign encodes the arrow of thermal processes.
The three slopes assemble into $\mathrm{d}S = \tfrac1T\mathrm{d}U + \tfrac PT\mathrm{d}V - \tfrac\mu T\mathrm{d}N$ , equivalently $\mathrm{d}U = T\mathrm{d}S - P\mathrm{d}V + \mu\mathrm{d}N$ .
Key idea: temperature, pressure, and chemical potential are the three first derivatives of the entropy $S(U,V,N)$ , defined by the second law's requirement that they equalize at equilibrium.

The core insight

Temperature, pressure, and chemical potential are the three first partial derivatives of the fundamental relation $S(U,V,N)$ : $1/T = (\partial S/\partial U)$ , $P/T = (\partial S/\partial V)$ , $-\mu/T = (\partial S/\partial N)$ . These definitions are forced by the second law — maximizing total entropy under exchange of a conserved quantity equalizes the matching derivative — and the concavity of $S$ in $U$ makes heat flow from hot to cold.

Where this lives today

The entropy-slope definition of temperature is not a philosophical nicety — it is the operationally correct definition used at the frontiers of physics, precisely where the naive "average kinetic energy" picture fails. In laser cooling and ultracold-atom experiments, temperatures of nanokelvin are defined and measured through the entropy–energy relationship of the trapped gas; the concept of a spin temperature, distinct from the motional temperature, is routine in NMR and in nuclear-magnetic experiments, and negative temperatures — impossible under the kinetic-energy definition — are now produced deliberately in cold-atom systems with bounded energy spectra, where $\partial S/\partial U < 0$ makes the system hotter than infinity. In semiconductor physics and electrochemistry, the chemical potential (the Fermi level) is the central quantity: it is what equalizes across a junction at equilibrium, what a voltage shifts, and what determines whether electrons flow. Battery voltages, solar-cell efficiencies, and transistor operation are all read off chemical-potential differences. In astrophysics and cosmology, the same entropy-derivative definitions describe systems from neutron-star matter to the primordial plasma. And conceptually, this article is where thermodynamic temperature ( $1/T = \partial S/\partial U$ ) is shown to be identical to the statistical temperature from $S = k_B\ln\Omega$ — the bridge that lets a single $T$ appear both in an engineer's Carnot engine and in a physicist's Boltzmann factor $e^{-E/k_BT}$ .

The physical picture

Forget mercury and thermometers for a moment and ask what happens when two bodies touch and share heat. Energy flows between them, but the total energy is fixed. The second law says the total entropy will climb to its maximum. Now, entropy is a function of each body's energy, $S_1(U_1)$ and $S_2(U_2)$ , and moving a sliver of energy $\mathrm{d}U$ from body 2 to body 1 changes the total by $\big[(\partial S_1/\partial U_1) - (\partial S_2/\partial U_2)\big]\mathrm{d}U$ . Entropy keeps rising until this bracket vanishes — until the two energy-slopes of entropy are equal. That common value at equilibrium is what we call temperature — more precisely, its reciprocal: $1/T = (\partial S/\partial U)$ . The definition is not arbitrary; it is chosen so that "thermal equilibrium" (equal slopes) coincides with "same temperature." A body whose entropy rises steeply as you add energy (large $\partial S/\partial U$ ) is cold; one whose entropy barely responds (small $\partial S/\partial U$ ) is hot. Feeding energy to a cold body buys a lot of entropy, so energy flows toward it.

Volume and particle number tell the same story with different exchanges. Let the two bodies also move a shared piston (exchange volume) at fixed total volume: total entropy peaks when $(\partial S/\partial V)_U$ matches, and identifying that with $P/T$ makes the condition "equal pressure" (mechanical equilibrium). Let them exchange particles through a permeable wall: the peak is where $(\partial S/\partial N)_U$ matches, i.e. $-\mu/T$ is equal — "equal chemical potential" (diffusive equilibrium). Three conserved exchanges, three intensive variables, three equilibrium conditions.

Figure

Three exchanges define three intensive variables via three entropy slopes

The core equations

The three intensive variables are the three first partial derivatives of the entropy in the entropy representation. Temperature is the inverse of the energy slope:

\frac{1}{T} = \left(\frac{\partial S}{\partial U}\right)_{V,N} \qquad (1)

where: $S$ — entropy; $U$ — internal energy; the derivative is at fixed $V, N$ . A steep slope (large $\partial S/\partial U$ ) means low temperature — adding energy buys a lot of entropy. This is the statistical-mechanical definition of temperature. The volume slope of entropy gives $P/T$ :

\frac{P}{T} = \left(\frac{\partial S}{\partial V}\right)_{U,N} \qquad (2)

where: $P$ — pressure. At fixed energy, entropy rises as the system expands (more accessible states), and the rate, weighted by $T$ , is the pressure. The particle-number slope gives $-\mu/T$ :

-\frac{\mu}{T} = \left(\frac{\partial S}{\partial N}\right)_{U,V} \qquad (3)

where: $\mu$ — chemical potential. Adding a particle usually lowers the entropy available per unit energy (the minus sign), and $\mu$ measures the energy cost of adding a particle at fixed $S, V$ . Assembling the three slopes gives the entropy-representation fundamental relation, and rearranging it recovers the energy representation:

\mathrm{d}S = \frac{1}{T}\,\mathrm{d}U + \frac{P}{T}\,\mathrm{d}V - \frac{\mu}{T}\,\mathrm{d}N \quad\Longleftrightarrow\quad \mathrm{d}U = T\,\mathrm{d}S - P\,\mathrm{d}V + \mu\,\mathrm{d}N \qquad (4)

where: the two forms are algebraically equivalent, related by solving one for $\mathrm{d}U$ . The entropy form makes the entropy-maximum principle natural; the energy form makes the energy-minimum principle natural. The equilibrium and stability conditions follow from maximizing total entropy under exchange. For thermal contact between two subsystems:

\left(\frac{\partial S_1}{\partial U_1}\right) = \left(\frac{\partial S_2}{\partial U_2}\right)\ \Rightarrow\ T_1 = T_2, \qquad \frac{\partial^2 S}{\partial U^2} < 0 \qquad (5)

where: the first-order condition gives thermal equilibrium ( $T_1 = T_2$ ); the second-order condition (concavity of $S$ ) makes it a genuine maximum, guarantees $C_V > 0$ , and forces heat to flow from hot to cold.

Derivation ATemperature as an entropy slope, and heat flow from the sign of the second derivative

The goal is to derive the definition of temperature (and the direction of heat flow) directly from the second law, showing that both the value condition (equal temperatures at equilibrium) and the approach condition (heat flows hot to cold) come from the first and second derivatives of the entropy. This is the article's owned result: not just that $1/T = \partial S/\partial U$ , but that the concavity $\partial^2 S/\partial U^2 < 0$ is what makes the equilibrium a stable maximum and dictates the arrow of thermal processes. We work it on two systems sharing energy and verify the sign symbolically.

Step 1. Two isolated systems, energies $U_1, U_2$ with $U_1 + U_2 = U_{\text{tot}}$ fixed, are brought into thermal contact. The total entropy is $S_{\text{tot}} = S_1(U_1) + S_2(U_2)$ ; the second law says it climbs to a maximum. Extremize: $\dfrac{\mathrm{d}S_{\text{tot}}}{\mathrm{d}U_1} = \dfrac{\partial S_1}{\partial U_1} - \dfrac{\partial S_2}{\partial U_2} = 0$ . Why: moving energy $\mathrm{d}U_1$ into system 1 removes it from system 2, so the two slopes appear with opposite signs.

Step 2. The stationarity condition is $(\partial S_1/\partial U_1) = (\partial S_2/\partial U_2)$ . Define this common equilibrium value as $1/T$ , so equal slopes means $T_1 = T_2$ — thermal equilibrium is equal temperature, the zeroth law recovered from entropy maximization. Why: identifying the equalized derivative with $1/T$ is exactly the definition that makes "thermal equilibrium" and "same temperature" the same statement.

Step 3. Check it is a maximum, not a saddle: $\dfrac{\mathrm{d}^2 S_{\text{tot}}}{\mathrm{d}U_1^2} = \dfrac{\partial^2 S_1}{\partial U_1^2} + \dfrac{\partial^2 S_2}{\partial U_2^2} < 0$ , which holds because entropy is concave in energy for each subsystem ( $\partial^2 S/\partial U^2 = -1/(U^2 C_V/\dots) < 0$ , equivalent to $C_V > 0$ ). For two ideal gases $S_i \propto \ln U_i$ , so $\partial^2 S_i/\partial U_i^2 = -a_i/U_i^2 < 0$ . Why: concavity of $S$ is the thermal-stability condition; without it the "equilibrium" would be unstable and the system would spontaneously separate hot from cold.

Step 4. Direction of heat flow: away from equilibrium, if $T_1 > T_2$ then $1/T_1 < 1/T_2$ , so $\mathrm{d}S_{\text{tot}}/\mathrm{d}U_1 = 1/T_1 - 1/T_2 < 0$ . Entropy therefore increases by decreasing $U_1$ — energy leaves the hotter body 1 and flows to the colder body 2. Heat flows hot → cold, forced by the second law and the concavity. Why: the sign of the first derivative away from the peak, guaranteed by the concave shape, is precisely the arrow of thermal processes. *Verified: the stationarity condition $1/T_1 = 1/T_2$ , the negative second derivative $\partial^2 S_{\text{tot}}/\partial U_1^2 < 0$ , and the hot→cold sign of $\mathrm{d}S_{\text{tot}}/\mathrm{d}U_1$ for a hotter body are all checked symbolically in verify_aq-sm-intensive-variables-entropy-derivatives.py.*

Go deeper · Derivation B · 1 minPressure and chemical potential from volume and particle exchange

Derivation BPressure and chemical potential from volume and particle exchange

The goal is to complete the trio by deriving $P/T$ and $-\mu/T$ as entropy slopes the same way temperature was derived — from maximizing total entropy under exchange of volume and of particles. This shows that all three intensive variables share one origin (the second law) and that the three equilibrium conditions (equal $T$ , equal $P$ , equal $\mu$ ) are three instances of one principle. We also confirm the ideal-gas forms so the abstract slopes become the familiar equation of state.

Step 1. Let two subsystems share a movable, diathermal piston, exchanging both energy and volume at fixed totals. Maximizing $S_{\text{tot}}$ over both $U_1$ and $V_1$ gives two conditions: $(\partial S_1/\partial U_1) = (\partial S_2/\partial U_2)$ and $(\partial S_1/\partial V_1) = (\partial S_2/\partial V_2)$ . Why: two independent conserved exchanges give two independent equilibrium conditions.

Step 2. The first gives $T_1 = T_2$ . Define the second common value as $P/T$ ; then $(\partial S_1/\partial V_1) = (\partial S_2/\partial V_2)$ combined with $T_1 = T_2$ gives $P_1 = P_2$ — mechanical equilibrium. Why: identifying the equalized volume-slope with $P/T$ makes "mechanical equilibrium" coincide with "equal pressure."

Step 3. For a permeable wall exchanging particles (and energy), maximizing over $N_1$ gives $(\partial S_1/\partial N_1) = (\partial S_2/\partial N_2)$ ; defining this as $-\mu/T$ yields $\mu_1 = \mu_2$ at equal $T$ — diffusive equilibrium. Particles flow from high $\mu$ to low $\mu$ , exactly as heat flows from high $T$ to low $T$ . Why: the chemical potential is the "pressure" for particle flow, and its gradient drives diffusion.

Step 4. Check the ideal gas: from the Sackur–Tetrode $S(U,V,N)$ , $(\partial S/\partial U) = \tfrac32 Nk_B/U$ gives $U = \tfrac32 Nk_BT$ ; $(\partial S/\partial V) = Nk_B/V$ gives $PV = Nk_BT$ . The abstract slopes reproduce the ideal-gas equation of state exactly. Why: this is the consistency check that the entropy-slope definitions are the same $T$ and $P$ that appear in the everyday gas law. *Verified: $(\partial S/\partial U) = \tfrac32 Nk_B/U \Rightarrow U = \tfrac32 Nk_BT$ and $(\partial S/\partial V) = Nk_B/V \Rightarrow PV = Nk_BT$ are checked symbolically; the numeric $T = 300\ \mathrm{K}$ for $1\ \mathrm{mol}$ at $U = 3740\ \mathrm{J}$ is confirmed in the verify script.*

Limiting cases

How the entropy slopes behave:

Case	Result	Meaning
Thermal contact	$(\partial S_1/\partial U_1) = (\partial S_2/\partial U_2) \Rightarrow T_1 = T_2$	Equal temperature = thermal equilibrium (zeroth law).
Ideal gas	$1/T = \tfrac32 Nk_B/U$ , $P/T = Nk_B/V$	Recovers $U = \tfrac32 Nk_BT$ and $PV = Nk_BT$ .
Concavity of $S$	$\partial^2 S/\partial U^2 < 0 \Leftrightarrow C_V > 0$	Stable maximum; heat flows hot → cold.
Bounded spectrum	$\partial S/\partial U$ can go negative	Negative temperature: $T < 0$ is hotter than $T = +\infty$ .
$T \to 0$	$\partial S/\partial U \to \infty$	Adding tiny energy hugely raises entropy near the ground state.
Diffusive contact	$(\partial S/\partial N)$ equal $\Rightarrow \mu_1 = \mu_2$	Particles flow high $\mu$ → low $\mu$ .

Common trap“Temperature is the average kinetic energy of the particles”

That is only true for a classical ideal gas, and it hides the real definition. The fundamental meaning of temperature is $1/T = (\partial S/\partial U)_{V,N}$ — how a system's entropy responds to added energy — and this definition works for every system: quantum gases, spin systems, radiation, systems with no kinetic energy at all. For a set of spins in a magnetic field there is no "average kinetic energy," yet temperature is perfectly well-defined through the entropy slope, and such systems can even reach negative temperature (a population inversion, where $\partial S/\partial U < 0$ ), which is hotter than any positive temperature. Equating temperature with kinetic energy also obscures why heat flows hot to cold — that comes from the sign of the second derivative of entropy, not from any statement about kinetic energy. Learn the entropy-slope definition; the kinetic-energy statement is a special case.

Worked examples

1 · Two bodies reach a common temperature by sharing energy (confidence builder)

Two identical blocks of the same material, one at $U_1 = 2$ (hot) and one at $U_2 = 1$ (cold) in units where each has entropy $S_i = \tfrac32\ln U_i$ (an ideal-gas-like form), are brought into contact with $U_1 + U_2 = 3$ fixed. Find the equilibrium split and confirm energy flows from hot to cold.

Step 1. Total entropy: $S_{\text{tot}} = \tfrac32\ln U_1 + \tfrac32\ln(3 - U_1)$ . Extremize: $\dfrac{\mathrm{d}S_{\text{tot}}}{\mathrm{d}U_1} = \dfrac{3}{2U_1} - \dfrac{3}{2(3-U_1)} = 0$ .

Step 2. Solve: $\dfrac{1}{U_1} = \dfrac{1}{3 - U_1} \Rightarrow U_1 = 3 - U_1 \Rightarrow U_1 = \mathbf{1.5}$ , so $U_2 = 1.5$ . Equal energies, hence equal temperatures ( $1/T \propto 1/U$ ) — the blocks equilibrate to the same temperature, as expected for identical bodies.

Step 3. Direction check at the initial hot state $U_1 = 2$ : $\dfrac{\mathrm{d}S_{\text{tot}}}{\mathrm{d}U_1}\Big|_{U_1=2} = \dfrac{3}{4} - \dfrac{3}{2} = -\mathbf{0.75} < 0$ . Entropy increases by decreasing $U_1$ — energy leaves the hotter block 1. Heat flows hot → cold, confirmed numerically.

Step 4. Stability: $\dfrac{\mathrm{d}^2 S_{\text{tot}}}{\mathrm{d}U_1^2} = -\dfrac{3}{2U_1^2} - \dfrac{3}{2(3-U_1)^2} < 0$ everywhere — the extremum is a genuine maximum, so the equilibrium is stable and the blocks stay equilibrated. The whole thermal behavior came from the first and second derivatives of entropy.

2 · The statistical temperature of a real gas (exam level)

One mole of monatomic ideal gas has internal energy $U = 3740\ \mathrm{J}$ . Using only the entropy-slope definition $1/T = (\partial S/\partial U)_{V,N}$ , find its temperature, then find the pressure at $V = 22.4\ \mathrm{L}$ from $(\partial S/\partial V)$ .

Step 1. From the Sackur–Tetrode entropy, $(\partial S/\partial U)_{V,N} = \tfrac32 Nk_B/U = \tfrac32 R/U$ (for one mole). So $1/T = \tfrac32(8.314)/3740 = \tfrac{12.47}{3740} = 3.335\times10^{-3}\ \mathrm{K^{-1}}$ .

Step 2. Invert: $T = 1/(3.335\times10^{-3}) = \mathbf{300\ K}$ . The temperature emerged purely from how fast entropy rises with energy — no thermometer, no "average kinetic energy," just a derivative of $S$ . (Equivalently $T = 2U/3R = 7480/24.94 = 300\ \mathrm{K}$ .)

Step 3. Pressure from the volume slope: $(\partial S/\partial V)_{U,N} = Nk_B/V = R/V$ , and $P = T(\partial S/\partial V)_{U,N} = TR/V$ . At $V = 22.4\ \mathrm{L}$ : $P = (300)(8.314)/0.0224 = 2494/0.0224 = \mathbf{1.11\times10^{5}\ Pa}$ ≈ 1 atm.

Step 4. Consistency: both intensive variables — temperature from $\partial S/\partial U$ and pressure from $\partial S/\partial V$ — came from the same entropy surface and reproduce the ideal-gas equation of state $PV = RT$ at standard conditions. This is the operational meaning of " $T$ and $P$ are entropy slopes": they are computable directly from $S(U,V,N)$ .

Origins · 1 minchemical potential and how this was discovered

chemical potential

concept · entropy slope for particle exchange

The chemical potential — the least intuitive of the three intensive variables — was introduced by Josiah Willard Gibbs in his 1875–78 memoir On the Equilibrium of Heterogeneous Substances as the quantity governing the flow of matter, exactly parallel to temperature governing heat flow and pressure governing volume change. Defined through the entropy slope $-\mu/T = (\partial S/\partial N)_{U,V}$ or equivalently $\mu = (\partial U/\partial N)_{S,V} = (\partial G/\partial N)_{T,P}$ , it is the energy cost of adding one particle to the system, and its gradients drive diffusion, osmosis, chemical reactions, and phase changes. Particles flow from high chemical potential to low, just as heat flows from high temperature to low. Gibbs's introduction of $\mu$ unified chemistry and thermodynamics: the condition for chemical and phase equilibrium is simply the equality of chemical potentials, and the entire edifice of physical chemistry — reaction equilibria, electrochemistry, membrane transport, semiconductor doping — rests on this single concept.

History

1848

— William Thomson (Kelvin) defines absolute temperature from Carnot's function, independent of substance.

1865

— Clausius writes dU = T dS − P dV, identifying T and P with entropy derivatives.

1875

— Gibbs introduces the chemical potential μ and the entropy-representation fundamental relation.

1877

— Boltzmann's S = k ln W gives the statistical footing for 1/T = ∂S/∂U.

1902

— Gibbs's Elementary Principles in Statistical Mechanics formalizes the ensemble definitions.

1951

— Purcell and Pound demonstrate negative spin temperature, dramatizing the ∂S/∂U definition.

Going further

References. Callen, Thermodynamics and an Introduction to Thermostatistics, 2nd ed., chs. 2–3 · Reif, Fundamentals of Statistical and Thermal Physics, ch. 3 (§3.3, the statistical definition of temperature) · Pathria & Beale, Statistical Mechanics, 3rd ed., ch. 1 · Kardar, Statistical Physics of Particles, ch. 4 · Kittel & Kroemer, Thermal Physics, ch. 2–3 (temperature and chemical potential) · Schroeder, An Introduction to Thermal Physics, ch. 3.

Provenance. The entropy-slope definitions $1/T = (\partial S/\partial U)$ , $P/T = (\partial S/\partial V)$ , $-\mu/T = (\partial S/\partial N)$ are standard (Callen chs. 2–3; Reif §3.3). The two-subsystem derivation of $T_1 = T_2$ , the concavity $\partial^2 S/\partial U^2 < 0$ , and the hot→cold heat-flow sign are checked symbolically. Numerical instances: two blocks $U_1 = U_2 = 1.5$ at equilibrium with $\mathrm{d}S_{\text{tot}}/\mathrm{d}U_1|_{U_1=2} = -0.75$ ; ideal-gas $T = 2U/3R = 300\ \mathrm{K}$ at $U = 3740\ \mathrm{J}$ and $P = RT/V = 1.11\times10^5\ \mathrm{Pa}$ . All in verify_aq-sm-intensive-variables-entropy-derivatives.py. $R = 8.314\ \mathrm{J\,mol^{-1}K^{-1}}$ .

Builds on: Second Law of Thermodynamics and Entropy · First Law of Thermodynamics and Internal Energy · Zeroth Law of Thermodynamics and Empirical Temperature · Work, Heat, and Inexact Differentials. Leads to: Thermodynamic Potentials and the Fundamental Relation · Legendre Transforms · Maxwell Relations · The Microcanonical Ensemble and the Postulate of Equal a Priori Probabilities.

Why is temperature the inverse of the entropy slope, 1/T, and not the slope itself?

It is a matter of definition, chosen to match everyday temperature and make the equilibrium condition read naturally. If you defined "temperature" as $\partial S/\partial U$ directly, then a hot body (which barely gains entropy per unit energy) would have a small value and a cold body a large one — backwards from intuition. Taking the reciprocal, $1/T = \partial S/\partial U$ , makes hot bodies have large $T$ and cold bodies small $T$ , and makes the ideal-gas result come out as the familiar $U = \tfrac32 Nk_BT$ with the right units. The factor also ensures that the energy-representation relation $\mathrm{d}U = T\,\mathrm{d}S + \dots$ has $T$ (not $1/T$ ) multiplying $\mathrm{d}S$ , matching $\delta Q = T\,\mathrm{d}S$ . So the inverse is not arbitrary — it is the choice that aligns the entropy-slope definition with the temperature you already know.

How can this definition give negative temperatures — isn't that below absolute zero?

Negative temperature is not "colder than absolute zero" — it is hotter than infinite temperature, and it is only possible for systems with a bounded energy spectrum. For an ordinary system, adding energy always increases entropy, so $\partial S/\partial U > 0$ and $T > 0$ . But consider a set of spins in a magnetic field: they have a maximum possible energy (all spins anti-aligned). As you approach that maximum, adding energy actually decreases the number of accessible states, so $\partial S/\partial U < 0$ and hence $T < 0$ . Such a population-inverted state will dump energy into any normal ( $T > 0$ ) system it contacts — which is why $T < 0$ is "hotter" than $T = +\infty$ . The temperature scale, ordered by hotness, runs $+0 \to +\infty \to -\infty \to -0$ . Lasers exploit exactly this population inversion. It is a striking confirmation that the entropy-slope definition, not the kinetic-energy picture, is the correct one.

What does the chemical potential actually "do" — I understand T and P but μ is fuzzy.

Think of it as the "temperature for particles." Just as temperature governs the flow of energy (heat moves from high $T$ to low $T$ ) and pressure governs the flow of volume (a piston moves toward lower $P$ ), the chemical potential governs the flow of particles: matter moves from high $\mu$ to low $\mu$ . Concretely, $\mu = (\partial U/\partial N)_{S,V}$ is the energy cost of adding one particle. When two regions can exchange particles, they reach equilibrium when their chemical potentials are equal — that is why a dissolved solute spreads until its concentration (and hence $\mu$ ) is uniform (diffusion), why water crosses a membrane toward higher solute concentration (osmosis, lower solvent $\mu$ ), and why electrons flow across a semiconductor junction until the Fermi levels ( $=\mu$ for electrons) align. Every "particles want to go somewhere" phenomenon is a chemical-potential gradient.

Do the three equilibrium conditions (equal T, P, μ) all have to hold at once?

Only if the two systems can exchange all three quantities. The rule is: for each conserved quantity the systems can exchange, the corresponding intensive variable equalizes. If a wall conducts heat but is rigid and impermeable, only energy is exchanged, so only $T_1 = T_2$ (thermal equilibrium) is required — the pressures and chemical potentials can differ. If the wall is also movable, volume is exchanged too and $P_1 = P_2$ is added. If it is also permeable, particles flow and $\mu_1 = \mu_2$ joins. So the equilibrium conditions are modular: each type of contact enforces its own equality. This is exactly why you can have, say, two gases at the same temperature but different pressures (separated by a rigid diathermal wall) sitting in stable equilibrium — a common and important situation.

Is the temperature defined here the same as the one from the partition function?

Yes, provably. This article defines $1/T = (\partial S/\partial U)_{V,N}$ from the entropy surface. Statistical mechanics computes that entropy from a microscopic model via $S = k_B\ln\Omega(U)$ (microcanonical) or derives the canonical ensemble with its Boltzmann factor $e^{-E/k_BT}$ . When you carry out the derivation of the canonical ensemble from a system in contact with a large reservoir, the parameter $\beta = 1/k_BT$ that appears in the Boltzmann factor is exactly $\beta = \partial\ln\Omega/\partial U = (1/k_B)(\partial S/\partial U)$ — the same $T$ as here. So the temperature that governs how entropy responds to energy (this article), the temperature in the Boltzmann factor, and the temperature of a Carnot engine are all the identical physical quantity. Demonstrating this equality is one of the central consistency results of statistical mechanics, and it is why "temperature" is unambiguous across the whole subject.

Why does entropy have to be concave in energy — what breaks if it isn't?

Concavity of $S(U)$ — meaning $\partial^2 S/\partial U^2 < 0$ — is the condition for thermal stability, and it is equivalent to a positive heat capacity, $C_V > 0$ . Here is what breaks without it. Suppose entropy were convex over some energy range (curving upward). Then splitting the system into two halves at different energies would raise the total entropy compared to keeping them equal — the uniform state would be unstable and the system would spontaneously develop hot and cold regions, which is absurd for a homogeneous body in equilibrium. Concavity guarantees that the entropy-maximum found by equalizing temperatures is a genuine, stable peak rather than a saddle or a minimum, and it ensures that adding energy always raises the temperature (positive $C_V$ ). Systems that violate concavity — most famously self-gravitating ones, which can have negative heat capacity — do not have a stable homogeneous equilibrium and require special non-extensive treatment. So concavity is not a technicality; it is what makes ordinary thermal equilibrium possible.