Thermodynamic Limit

A thermodynamic limit is a controlled sequence of larger quantum systems in which selected densities and local observables approach well-defined bulk behavior. It is not shorthand for “a very large number,” and it is not determined by $N\to\infty$ alone.

For particles in a region of volume $V$, the standard fixed-density limit is

$$\begin{gathered} N\longrightarrow\infty, \qquad V\longrightarrow\infty, \\ \frac{N}{V} \longrightarrow n \quad \text{with }0<n<\infty. \end{gathered}$$

For a lattice, one instead takes a sequence of finite regions $\Lambda_L$ whose number of sites tends to infinity:

$$|\Lambda_L| \longrightarrow \infty.$$

A complete statement also specifies:

• the shape and spatial dimension of the regions;
• the boundary conditions;
• which densities or couplings are held fixed;
• how long-range interactions are scaled;
• which observable, free-energy density, or correlation function is expected to converge;
• the order relative to source, time, frequency, temperature, continuum, and other limits.

This page gives the canonical physics-level introduction. Rigorous existence theorems, quasilocal operator algebras, KMS states, and classification of infinite-volume Gibbs states require a separate mathematical treatment.

Why Take an Infinite-System Limit?

No laboratory sample contains infinitely many particles. The thermodynamic limit is useful because it separates bulk structure from finite-size details.

It can:

• turn extensive quantities into stable densities;
• suppress ordinary boundary-to-volume corrections;
• produce continuous spectra from increasingly dense finite-size levels;
• define phases by stable bulk properties;
• permit nonanalytic free energies and sharp thermal transitions;
• support exact long-range order and distinct symmetry-breaking states;
• expose universal scaling independent of many microscopic details.

The limit is an organizing idealization. Its value comes from controlled convergence of observables measured in large finite systems.

A Thermodynamic Sequence

Consider a $d$-dimensional cubic lattice with linear size $L$:

$$\Lambda_L = \{1,\ldots,L\}^{d}, \qquad |\Lambda_L| = L^d.$$

The boundary has order

$$|\partial\Lambda_L| \sim L^{d-1},$$

so

$$\frac{ |\partial\Lambda_L| }{ |\Lambda_L| } \sim \frac{1}{L} \longrightarrow 0.$$

This is why surface corrections often become subextensive for short-range systems in regular shapes.

The statement is not automatic for arbitrary sequences. A highly perforated region can have boundary proportional to volume. A long thin strip can retain effectively lower-dimensional behavior. One should specify a regular sequence whose boundary-to-volume ratio vanishes.

In rigorous lattice statistical mechanics, such sequences are often formulated as van Hove or Følner-type sequences. The terminology matters less here than the physical condition: the bulk must grow faster than the boundary.

Bulk Limits of Extensive Quantities

The scaling classification is made before the limit is claimed. Extensive and Intensive Quantities owns the definitions, normalization ledger, boundary hierarchy, and distinction between extensivity and additivity.

$$X_L = x\,|\Lambda_L| + o(|\Lambda_L|)$$

is a candidate extensive form. The thermodynamic-limit question is whether the density

$$\lim_{L\to\infty} \frac{X_L}{|\Lambda_L|}$$

exists for the declared sequence and whether it is stable under the allowed boundary and shape choices. A bound $X_L=O(|\Lambda_L|)$ alone does not establish convergence. Subextensive surface, edge, corner, topological, and logarithmic pieces may vanish in the density while remaining important observables.

Local Observables Are Often the Natural Targets

An infinite system does not generally possess a normalizable state vector in one finite tensor-product Hilbert space. The physically stable questions are often local.

Let $A_X$ be an observable supported in a fixed finite region $X$. If $\rho_L$ is a state on a larger region $\Lambda_L\supset X$, one asks whether

$$\lim_{L\to\infty} \operatorname{Tr} \left( \rho_L A_X \right)$$

exists.

The region $X$ is held fixed while the environment around it grows. This differs from asking for convergence of the global density matrix in trace norm, which is usually neither available nor necessary.

Correlation functions can be treated similarly:

$$\lim_{L\to\infty} \langle A_xB_y\rangle_L$$

at fixed $x$ and $y$, followed when appropriate by a large-separation limit

$$|x-y| \longrightarrow \infty.$$

The thermodynamic and large-distance limits are distinct operations.

Stability and Short-Range Structure

A conventional thermodynamic limit requires the energy not to decrease faster than linearly with particle number. A standard stability condition has the form

$$E_{N,V} \geq -B N$$

for a constant $B$ independent of $N$ and $V$.

For sufficiently local or rapidly decaying interactions, widely separated bulk regions interact weakly enough that surface corrections do not overwhelm volume scaling.

These conditions should not be treated as automatic. Singular attractive interactions can produce collapse. Slowly decaying long-range forces can make the energy superextensive. Charged systems require neutrality, screening, or more careful limiting procedures.

The rigorous conditions depend on the model. At the physics level, always ask:

1. Is the Hamiltonian bounded below in a way compatible with increasing size?
2. Does the interaction energy scale proportionally to volume?
3. Do distant regions couple weakly enough for a bulk density to emerge?

Long-Range and All-to-All Interactions

Long-range interactions make the sequence itself part of the model. Extensive and Intensive Quantities derives the complete-graph and power-law counting diagnostics and shows why Kac normalization can restore an order-volume energy without restoring locality or additivity.

For the limiting problem, the essential point is that different size-dependent normalizations define different Hamiltonian sequences. A normalization should therefore be stated and physically justified, not inserted mechanically. Spatial power laws, neutral Coulomb matter, gravitational models, and engineered collective couplings have distinct stability and convergence questions.

Finite Positive-Temperature Systems Are Analytic

Let $H_L(g)$ be a finite-dimensional Hamiltonian depending analytically on a real coupling $g$. At finite inverse temperature $0<\beta<\infty$,

$$Z_L(\beta,g) = \operatorname{Tr} e^{-\beta H_L(g)}$$

is positive for real $g$. The finite-size free energy

$$F_L(\beta,g) = -\frac{1}{\beta} \ln Z_L(\beta,g)$$

is therefore analytic under the stated finite-dimensional and regularity assumptions.

Avoid overgeneralizing this claim:

• it concerns finite positive temperature;
• the Hilbert space or partition function must be well defined;
• zero temperature is the additional limit $\beta\to\infty$;
• a finite system can have exact level crossings and nonanalytic ground-state energy at zero temperature;
• complex zeros of $Z_L$ can approach the real axis as size grows.

Sharp thermal phase-transition singularities arise after an infinite-system or another singular limit, not from replacing a finite crossover by rhetoric.

How Analytic Finite Systems Produce a Nonanalytic Limit

A sequence of analytic functions can converge to a nonanalytic function. The elementary example

$$f_L(h) = -\frac{1}{L} \ln \left[ 2\cosh(Lh) \right]$$

is analytic for every finite $L$. For fixed nonzero $h$,

$$\lim_{L\to\infty} f_L(h) = -|h|.$$

The limiting function has a cusp at $h=0$.

The example is a mathematical illustration, not a complete local many-body model. It shows why finite-size analyticity does not forbid a bulk singularity: convergence need not preserve analyticity uniformly near the singular point.

In statistical mechanics, partition-function zeros can accumulate toward the physical parameter axis, correlation lengths can diverge, and increasingly sharp finite-size crossovers can converge to a genuine phase transition.

What Defines a Phase?

A phase is not merely a region where one finite-size plot looks smooth. Depending on context, a thermodynamic phase can be characterized by:

• analyticity and nonanalytic boundaries of thermodynamic potentials;
• symmetry and order parameters;
• correlation decay and long-range order;
• topological or other nonlocal structure;
• response coefficients;
• excitation structure and spectral gaps;
• the set of infinite-volume equilibrium states.

No single item is universal in every setting. For example, topological phases need not be distinguished by a local symmetry-breaking order parameter. A first-order transition and a continuous transition have different finite-size signatures.

The thermodynamic limit supplies the arena in which these distinctions become sharp. Phases of Matter in Many-Body QM develops the general phase concept and the complementary diagnostic fingerprint.

Boundary Conditions

Common finite-system choices include:

• open boundaries;
• periodic boundaries;
• antiperiodic or twisted boundaries;
• fixed boundary spins;
• boundary fields;
• hard-wall or periodic continuum boxes.

For short-range systems in regular geometries, the bulk free-energy density often becomes independent of ordinary boundary conditions:

$$\lim_{L\to\infty} \frac{ F_L^{\mathrm{open}} - F_L^{\mathrm{periodic}} }{ |\Lambda_L| } = 0.$$

This does not mean boundaries are physically irrelevant.

• Surface free energies can remain order $L^{d-1}$.
• Edge states can survive at low energy.
• Boundary fields can select a symmetry-breaking phase.
• Twisted boundaries can diagnose stiffness or topology.
• Frustrated and long-range systems can retain stronger shape dependence.
• Finite-size spectra can depend sensitively on boundary sectors.

Always distinguish “the bulk density is boundary independent” from “every observable is boundary independent.”

Shape and Aspect Ratio

Taking $L_x,L_y,L_z\to\infty$ can produce different behavior if aspect ratios are not controlled.

A three-dimensional box with

$$\frac{L_x}{L_y}, \quad \frac{L_y}{L_z}$$

held finite approaches a three-dimensional bulk. If $L_x\to\infty$ while $L_y$ and $L_z$ remain fixed, the sequence approaches a quasi-one-dimensional system.

Dimensional crossover is physical, not a technical nuisance. A thermodynamic-limit statement must say which lengths grow and how.

Order of Limits

Two limits commute only if the model and observable justify exchanging them. Many central many-body phenomena are defined precisely by noncommuting limits.

Source and volume

Let a field $h$ couple to an order parameter $M$:

$$H_{L,h} = H_L-hM.$$

A symmetry-selected value can be defined schematically by

$$m_+ = \lim_{h\to0^+} \lim_{L\to\infty} \frac{ \langle M\rangle_{L,h} }{ |\Lambda_L| }.$$

Taking $h\to0$ first can restore the finite-system symmetry and give zero. The canonical phase-selection interpretation belongs in Spontaneous Symmetry Breaking.

Time and volume

A finite isolated quantum system with a discrete spectrum can exhibit recurrences. A literal pointwise limit

$$\lim_{t\to\infty} \langle A(t)\rangle_L$$

may not exist.

One may instead use a long-time average, a large-system limit before the long-time limit, or an observational time window:

$$\lim_{t\to\infty} \lim_{L\to\infty} \langle A(t)\rangle_L.$$

These procedures are not interchangeable by default. Thermalization claims must state which is meant.

Nonequilibrium Overview turns this warning into a full time-window and evidence ledger, including dephasing, recurrence, transport, prethermal, and heating scales.

Frequency and wave number

Response functions can have distinct static and transport limits:

$$\lim_{\omega\to0} \lim_{\mathbf q\to0} \chi(\mathbf q,\omega) \neq \lim_{\mathbf q\to0} \lim_{\omega\to0} \chi(\mathbf q,\omega).$$

The difference can separate equilibrium susceptibility, compressibility, stiffness, and transport response. The operator definitions and physical perturbation must accompany the limit. Fluctuations and Susceptibilities develops the equilibrium variance scaling and critical enhancement behind these static coefficients. Susceptibilities gives the named source, units, and local-versus-uniform conventions.

Superfluidity in Condensed Matter applies this limit-order contract to neutral stiffness and film inference; this page retains the general thermodynamic-limit rules.

Temperature and volume

A quantum critical point often involves

$$T\longrightarrow0$$

together with $L\to\infty$. A finite-size gap can remain nonzero at every finite $L$ and close only in the bulk:

$$\Delta_L \longrightarrow 0.$$

Taking $T\to0$ at fixed size isolates a finite spectrum. Taking $L\to\infty$ first can expose gapless bulk excitations and singular response.

The Transverse-Field Ising Model gives a concrete example in which finite-size symmetry splitting, the bulk quasiparticle gap, and thermodynamic symmetry breaking must be kept distinct.

Continuum and volume

The continuum limit sends a lattice spacing $a\to0$ while physical lengths and renormalized parameters are controlled. The thermodynamic limit sends physical volume to infinity. They answer different questions:

$$a\to0 \qquad \text{versus} \qquad L_{\mathrm{phys}}\to\infty.$$

A calculation may require both, but it must state their order and scaling relation. Real-Space Representation gives the operator-normalization and coupling-scaling dictionary for a spatial lattice approaching a continuum.

Finite Systems Approximate the Limit Observable by Observable

There is no universal size above which a system becomes thermodynamic. The relevant criterion compares $L$ with the physical scales controlling an observable.

If the correlation length is $\xi$, bulk local behavior often converges when

$$L\gg\xi.$$

Near a continuous transition,

$$\xi \sim |g-g_c|^{-\nu}$$

can become comparable to or larger than every accessible finite size. Finite-size effects then carry the critical information rather than merely contaminating it.

Other relevant scales include:

• thermal wavelength;
• mean free path;
• screening length;
• localization length;
• entanglement length;
• inverse spectral gap;
• recurrence time;
• boundary penetration depth.

A system can be thermodynamic for one local observable and strongly finite-size dependent for another.

Finite-Size Scaling Preview

Near a continuous transition, a finite system cannot support a truly divergent correlation length; size becomes an infrared cutoff. This is one mechanism among several. Finite-Size Effects owns the physical audit of quantized modes, gap identities, boundaries, shells, commensurability, correlation cutoffs, traps, and recurrence windows.

Once the mechanism and comparison limit are clear, Critical Exponents and Scaling supplies continuous critical ansätze, correction terms, crossing drift, and scaling functions. Finite-Size Scaling in Numerics owns fit windows, covariance, competing models, data collapse, and robust extrapolation. Finite-Temperature Phase Transitions specializes the logic to thermal singularities and transition order.

Ensemble Equivalence Is Conditional

Microcanonical, canonical, and grand-canonical ensembles impose different constraints. In suitable short-range systems away from problematic coexistence regions, they can agree for local bulk observables in the thermodynamic limit.

That equivalence is not automatic.

It can fail or require qualification for:

• finite systems;
• long-range nonadditive interactions;
• phase coexistence;
• constrained or nonergodic sectors;
• observables sensitive to global fluctuations;
• negative-heat-capacity regimes in nonadditive systems;
• integrable dynamics and generalized ensembles.

The thermodynamic limit does not erase the assumptions defining an ensemble.

Ensemble Equivalence owns the supporting-line, large-deviation, local-state, coexistence, and nonequivalence criteria. This page retains ownership of how the finite systems, boundary conditions, and scaling variables are taken to infinity.

Infinite-Volume States Need Not Be Density Matrices

For finite $\Lambda_L$, an equilibrium state can be written

$$\rho_L = \frac{ e^{-\beta H_L} }{ Z_L }.$$

In an infinite system, the formal trace

$$\operatorname{Tr} e^{-\beta H_\infty}$$

need not exist. A global Gibbs density matrix may therefore be the wrong object.

The algebraic formulation instead describes a state as a positive normalized functional $\omega$ on a quasilocal observable algebra:

$$\omega(I)=1, \qquad \omega(A^\dagger A)\geq0.$$

Thermal equilibrium can be encoded by the KMS condition. Distinct phases can correspond to distinct infinite-volume states or inequivalent Hilbert-space representations.

This page does not develop that machinery. It cross-links to the Algebraic Formulation Overview so that the finite-volume density-matrix notation is not extrapolated beyond its domain without warning.

Several Meanings of Large N

The thermodynamic limit is only one large-system limit.

Limit	Quantity made large	Typical quantity held fixed
Particle thermodynamic limit	$N,V$	density $N/V$
Lattice thermodynamic limit	number of sites	couplings and shape class
Mean-field collective limit	number of interacting units	interaction rescaled for extensivity
Large-$N$ internal-symmetry limit	number of field components or flavors	model-dependent rescaled coupling
Semiclassical limit	action relative to $\hbar$	classical scales
Continuum limit	number of lattice points	physical size and renormalized parameters

These limits can coexist, but none should be substituted for another without an explicit mapping.

Large-N and Saddle-Point Methods Preview develops the internal-component limit and explains why it is distinct from increasing particle number or spatial volume.

Worked Example: Independent Spins

Consider $L$ noninteracting spin-$1/2$ moments in a field:

$$H_L = -h \sum_{i=1}^{L} \sigma_i^z.$$

The partition function factorizes:

$$Z_L = \left[ 2\cosh(\beta h) \right]^L.$$

The free energy per site is already independent of $L$:

$$f_L = -\frac{1}{\beta L} \ln Z_L = -\frac{1}{\beta} \ln \left[ 2\cosh(\beta h) \right].$$

Therefore

$$f = \lim_{L\to\infty}f_L$$

exists and is analytic for finite $\beta$ and real $h$.

Large system size alone does not create a phase transition. Interactions, dimensionality, symmetry, and fluctuation structure determine whether singular bulk behavior occurs.

Worked Example: Surface Corrections

Suppose a short-range $d$-dimensional system has

$$F_L = f_{\mathrm b}L^d + f_{\mathrm s}L^{d-1} + o(L^{d-1}).$$

Dividing by volume gives

$$\frac{F_L}{L^d} = f_{\mathrm b} + \frac{f_{\mathrm s}}{L} + o(L^{-1}).$$

The bulk free-energy density approaches $f_{\mathrm b}$, while the surface free energy remains a meaningful subleading observable.

A Thermodynamic-Limit Checklist

Before accepting a bulk claim, ask:

1. Sequence: Which finite systems approach infinity?
2. Geometry: Which dimensions and aspect ratios grow?
3. Density: Which particle, energy, charge, or filling densities are fixed?
4. Interaction scaling: Is the energy extensive?
5. Boundary conditions: Which are used, and what can depend on them?
6. Observable: Is it local, intensive, extensive, or boundary sensitive?
7. Convergence: What finite-size evidence supports the limit?
8. Order: Which other limits are taken before or after size?
9. Ensemble: Which constraints define the state?
10. Regularity: Is existence of the partition function or bulk state established or assumed?
11. Knowledge status: Is the result exact, rigorous, asymptotic, numerical, or heuristic?

Canonical Boundaries

This page owns the non-rigorous physical meaning of the thermodynamic limit, its required specifications, and its central limit-order caveats.

Related canonical homes retain narrower subjects:

• spontaneous symmetry breaking owns source-selected order parameters and symmetry sectors;
• Finite-Size Effects owns the physical mechanism audit, while the critical-scaling and numerical pages own detailed ansätze, estimators, and corrections;
• ensemble pages own derivations of microcanonical, canonical, and grand-canonical states;
• open-systems pages own bath-induced relaxation and thermodynamic processes;
• mathematical and algebraic treatments own existence theorems, KMS states, and infinite-volume operator algebras;
• QFT treatments own continuum renormalization and field-theoretic infrared limits in full depth.

Common Mistakes

• Writing $N\to\infty$ without specifying volume or density.
• Taking volume to infinity while accidentally diluting the system to zero density.
• Ignoring geometry, aspect ratio, or boundary-to-volume scaling.
• Assuming the thermodynamic limit exists for every interaction.
• Forgetting stability or extensivity for long-range models.
• Calling a sharp finite-size crossover a phase transition without scaling evidence.
• Claiming finite positive-temperature nonanalyticity for a regular finite-dimensional partition function.
• Forgetting that zero temperature is another singular limit.
• Exchanging source, time, frequency, wave-number, continuum, or volume limits without justification.
• Assuming bulk boundary independence implies that edge observables vanish.
• Treating ensemble equivalence as automatic.
• Writing a normalized global Gibbs density matrix for an infinite system without checking whether the trace exists.
• Using one system-size threshold for every observable.
• Confusing a large-$N$ flavor expansion with the particle thermodynamic limit.

Exercises

Boundary-to-volume scaling

For a $d$-dimensional hypercubic region with side length $L$, show that the fraction of sites within a fixed distance $r$ of the boundary vanishes as $L\to\infty$.

Solution

The total number of sites is order $L^d$. For fixed $r$, the boundary layer has thickness independent of $L$ and volume of order

$$rL^{d-1}.$$

Therefore the fraction scales as

$$\frac{ rL^{d-1} }{ L^d } \sim \frac{r}{L} \longrightarrow 0.$$

The conclusion assumes $r$ remains fixed and the region grows regularly in every spatial direction.

Analytic sequence with a cusp

For

$$f_L(h) = -\frac{1}{L} \ln \left[ 2\cosh(Lh) \right],$$

show that the pointwise large-$L$ limit is $-|h|$.

Solution

Use

$$2\cosh(Lh) = e^{L|h|} \left( 1+e^{-2L|h|} \right).$$

Then

$$\begin{aligned} f_L(h) ={}& -|h| \\ &- \frac{1}{L} \ln \left( 1+e^{-2L|h|} \right). \end{aligned}$$

The second term vanishes as $L\to\infty$, including at $h=0$, where it equals $-(\ln2)/L$. Thus

$$\lim_{L\to\infty}f_L(h) = -|h|.$$

Every finite-$L$ function is analytic, but the pointwise limit has a cusp.

Noncommuting source and size limits

Use the toy finite-size order parameter

$$m_L(h) = \tanh(Lh)$$

to compare

$$\lim_{h\to0^+} \lim_{L\to\infty} m_L(h)$$

with

$$\lim_{L\to\infty} \lim_{h\to0} m_L(h).$$

Solution

At fixed $h>0$,

$$\lim_{L\to\infty} \tanh(Lh) = 1.$$

Removing the source afterward gives

$$\lim_{h\to0^+} \lim_{L\to\infty} m_L(h) = 1.$$

At any finite $L$,

$$\lim_{h\to0} \tanh(Lh) = 0,$$

so

$$\lim_{L\to\infty} \lim_{h\to0} m_L(h) = 0.$$

The toy function illustrates noncommuting limits. Establishing spontaneous symmetry breaking in a physical model requires the model’s state structure and observables, not this analogy alone.

Extract a dynamical exponent

At a candidate quantum critical point, numerical gaps satisfy

$$\frac{\Delta_{2L}}{\Delta_L} \longrightarrow \frac{1}{4}.$$

Assuming $\Delta_L\sim L^{-z}$, determine $z$ and state one reason the conclusion might fail at accessible sizes.

Solution

The scaling ansatz gives

$$\frac{\Delta_{2L}}{\Delta_L} \sim \frac{(2L)^{-z}}{L^{-z}} = 2^{-z}.$$

Therefore

$$2^{-z} = \frac{1}{4} = 2^{-2},$$

so $z=2$.

At accessible sizes, irrelevant operators, boundary conditions, aspect ratios, crossover scales, or an incorrectly tuned coupling can produce substantial corrections. A stable estimate requires several sizes and a correction-to-scaling analysis.

Cross-Links

• Connected Correlation Functions
• Structure Factors
• Susceptibilities
• Many-Body and Quantum Statistical Mechanics
• Microcanonical Ensemble
• Why Many-Body Physics Is Different
• Core Objects and Notation
• Scaling of Hilbert Space
• Thermal Density Operators
• Canonical Ensemble
• Ensemble Equivalence
• Thermodynamic Potentials
• Quantum Phase Transitions
• Transverse-Field Ising Model
• XXZ Spin Chain
• Statistical Mechanics Checklist
• Spontaneous Symmetry Breaking Preview
• Algebraic Formulation Overview
• Density of States
• Quantum Thermodynamics
• Condensed-Matter Roadmap

References

• R. K. Pathria and P. D. Beale, Statistical Mechanics, 3rd ed., Elsevier (2011).
• M. Kardar, Statistical Physics of Particles, Cambridge University Press (2007).
• L. D. Landau and E. M. Lifshitz, Statistical Physics, Part 1, 3rd ed., Butterworth–Heinemann (1980).
• N. Goldenfeld, Lectures on Phase Transitions and the Renormalization Group, CRC Press (2018).
• S. Sachdev, Quantum Phase Transitions, 2nd ed., Cambridge University Press (2011).
• D. Ruelle, Statistical Mechanics: Rigorous Results, World Scientific (1999).
• B. Simon, The Statistical Mechanics of Lattice Gases, Volume I, Princeton University Press (1993).
• O. Bratteli and D. W. Robinson, Operator Algebras and Quantum Statistical Mechanics 2, 2nd ed., Springer (1997).