Finite-Size Effects

A finite quantum system is not a defective infinite system. Its discrete spectrum, boundaries, shell structure, edge modes, and recurrence times are physical. The same features can also obstruct a bulk inference when they are mistaken for thermodynamic behavior.

This page provides an observable-specific finite-size audit. Given a calculation or experiment, its job is to identify why the finite result differs from a stated bulk reference, decide whether that difference is negligible, diagnostic, or the phenomenon of interest, and route any extrapolation to the appropriate canonical treatment.

The central rule is:

“Finite-size effects are small” has no meaning until the observable, geometry, state, resolution, and comparison scale are named.

Helpful background. Extensive and Intensive Quantities fixes normalization and subleading terminology. Thermodynamic Limit explains how a family of finite systems defines a bulk claim. Neither is required to understand a finite system as a physical object.

Finite Size Is an Observable-Relative Comparison

For a normalized observable with a well-defined bulk reference, write

$$\delta O_L^{\mathrm{fs}} = O_L-O_\infty.$$

This notation is incomplete until the family of Hamiltonians, states, geometries, boundaries, and normalizations is fixed. It is inapplicable when no comparison limit is intended—for example, when an edge spectrum or a finite trapped cloud is itself the target.

Different mechanisms may contribute schematically,

$$\begin{aligned} \delta O_L^{\mathrm{fs}} \sim{}& \delta O_L^{\mathrm{disc}} + \delta O_L^{\mathrm{boundary}} + \delta O_L^{\mathrm{shell}} \\ &+ \delta O_L^{\mathrm{corr}} + \delta O_L^{\mathrm{constraint}} +\cdots. \end{aligned}$$

This is a diagnostic ledger, not a universal additive expansion. The mechanisms can interfere: a boundary condition changes the allowed momenta, which changes shell filling, which can move the lowest state into another symmetry sector.

“Size” may mean a linear extent $L$, a volume $V$, a number of sites, a particle number $N$, a trap radius, a graph diameter, or a time window. More than one may matter.

• $L/\xi_O$ — correlation containment. Can the system contain the correlations relevant to $O$? If $L\lesssim\xi_O$, the boundary or infrared cutoff reaches the observable.
• $d_{\partial}/\xi_O$ — boundary isolation. Is a local probe far from the nearest boundary? A central value can be bulk-like while an edge value is not.
• Mode spacing versus $k_{\mathrm B}T$ — thermal resolution. A continuum density-of-states approximation can be inaccurate when only a few modes are thermally active.
• Mode spacing versus probe resolution — spectral resolution. Unresolved lines may look continuous without being irreversible.
• $t_{\mathrm{obs}}/t_{\mathrm{return}}$ — dynamical window. Has information reached a boundary and returned? Late-time behavior may be a wraparound effect.
• $L/\lambda_O$ — commensurability. Does the geometry fit the structure being tested? Odd–even or shape oscillations can dominate a short sequence.

Here $\xi_O$ is the length relevant to the declared observable, $d_{\partial}$ is its distance from a physical boundary, and $\lambda_O$ is an ordering wavelength or another spatial period. There need not be one universal correlation length for every question.

A finite-size feature can play three different roles:

1. Correction: a boundary contribution biases an estimate of a bulk energy density.
2. Diagnostic: a gap or crossing drifts with size in a way that distinguishes candidate bulk regimes.
3. Target physics: an edge state, mesoscopic shell, quantum dot spectrum, or trapped cloud is interesting precisely because the system is finite.

The role must be declared before “removing” the effect.

Physical Finite Size Is Not Numerical Error

Four error classes are often mixed together:

• Physical finite-size structure: a discrete momentum grid or open-edge mode changes when the physical size, shape, or boundary condition changes.
• Numerical error: an eigensolver residual or tensor-network truncation changes when the algorithmic tolerance or representation is tightened.
• Experimental resolution: a linewidth, finite imaging scale, or limited observation time changes when the apparatus or its response model changes.
• Model discrepancy: omitted interactions, disorder, or trap inhomogeneity changes only when the physical model and calibration improve.

Exact Diagonalization can determine an eigenvalue of a declared finite Hamiltonian to nearly machine precision. That removes solver error from that eigenvalue; it does not remove the Hamiltonian’s physical finite-size dependence. Conversely, a laboratory edge or shell feature is not an “error” merely because it disappears in a bulk limit.

This distinction should appear in every uncertainty statement. A result can be numerically converged and physically far from the bulk, or physically bulk-like while carrying substantial measurement uncertainty.

Discrete Momenta and Spectral Resolution

For a translation-invariant ring of circumference $L$, periodic boundary conditions give

$$k_n = \frac{2\pi n}{L}, \qquad \Delta k = \frac{2\pi}{L}.$$

Suppose a low-energy branch near $k_0$ behaves as

$$\epsilon(k_0+q)-\epsilon(k_0) \simeq A\lvert q\rvert^z.$$

Here $A$ fixes the low-energy scale and $z>0$ is the dispersion exponent in the stated regime.

If the first allowed excitation has $\lvert q\rvert=2\pi/L$, its quantization gap scales as

$$\Delta_L \simeq A\left(\frac{2\pi}{L}\right)^z.$$

A linear branch gives $L^{-1}$; a quadratic minimum gives $L^{-2}$. Boundary conditions, conservation laws, and sector restrictions can change the smallest allowed $q$, so the coefficient and sometimes the accessible sequence must be stated.

Three spacings must not be conflated:

• one-particle mode spacing: the separation of allowed orbital or quasiparticle energies;
• low-energy excitation gap: the energy difference between specified many-body states near the ground state or another reference state;
• mean many-body level spacing: the average separation among many eigenvalues in a declared energy window and symmetry sector.

The last quantity can be written schematically as

$$\delta_{\mathrm{MB}}(E) \sim \frac{1}{\rho_{\mathrm{MB}}(E)}.$$

At finite entropy density, the many-body density of states $\rho_{\mathrm{MB}}$ is often exponential in volume. The mean spacing may therefore be exponentially small even while a physically selected low-energy excitation remains gapped. A tiny global mean spacing is not evidence that the bulk phase is gapless.

A spectrum looks continuous only relative to a resolution. Compare the relevant spacing separately with $k_{\mathrm B}T$, a physical linewidth, an instrumental bandwidth, and the Fourier resolution $\hbar/t_{\mathrm{obs}}$. Artificial broadening used to plot discrete lines is a visualization or regularization choice; by itself it is not a physical lifetime.

“The Finite-Size Gap” Is Ambiguous

A quoted gap is not reproducible until both states, their quantum numbers, the boundary condition, and the size sequence are specified.

• Bulk excitation gap: $\Delta_L\to\Delta_\infty>0$ asks whether the infinite system is gapped in the chosen sector.
• Critical quantization gap: behavior such as $\Delta_L\propto L^{-z}$ probes the low-energy dispersion and dynamic scaling at a critical point.
• Symmetry-partner tunneling: a model-dependent form such as $L^p e^{-cL^d}$ can signal finite symmetric combinations approaching distinct broken-symmetry states.
• Edge-mode hybridization: behavior such as $e^{-L/\xi_{\mathrm e}}$ can arise when modes localized on opposite boundaries overlap.
• Anderson tower scale: an inverse-volume family can organize a finite symmetric spectrum toward continuous symmetry breaking.
• Shell or commensurability gap: oscillations with parity, shape, or filling test whether the finite momentum grid fits the occupied or ordered structure.
• First-order avoided crossing: an exponentially small, geometry-dependent splitting can describe tunneling between competing finite-volume phases.
• Mean sector level spacing: a rapidly densifying spectrum controls spectral resolution and long-time scales within the declared symmetry and energy window.

These patterns are clues, not unique identifiers. An exponential-looking sequence over three small sizes does not prove an edge mode, and a $1/L$ gap can arise from more than one mechanism. State overlaps, spatial profiles, symmetry labels, boundary dependence, and operator matrix elements identify what the levels are.

Spontaneous Symmetry Breaking owns cat-state splittings and towers of states. The Transverse-Field Ising Model provides a model-specific comparison among symmetry splitting, quasiparticle gaps, boundaries, and criticality.

Boundaries, Shape, and Commensurability

An open system can have a thin boundary region whose local physics differs by order one from the bulk even though its fraction of the total volume vanishes. This distinction separates local convergence from spatially averaged convergence.

Consider a one-dimensional observable with a schematic boundary correction

$$\begin{aligned} \delta o_L(x) ={}& A e^{-x/\xi} + A e^{-(L-x)/\xi}, \\ &0\le x\le L. \end{aligned}$$

At the center,

$$\delta o_L(L/2) = 2A e^{-L/(2\xi)},$$

whereas its spatial average is

$$\begin{aligned} \overline{\delta o}_L &= \frac{1}{L} \int_0^L \delta o_L(x)\,dx \\ &= \frac{2A\xi}{L} \left(1-e^{-L/\xi}\right). \end{aligned}$$

For $L\gg\xi$, the center is exponentially close to the bulk while the whole-system average retains a $1/L$ boundary fraction. More generally, a boundary layer of thickness $\xi$ in a $d$-dimensional box occupies the fraction

$$1- \left(1-\frac{2\xi}{L}\right)^d \simeq \frac{2d\xi}{L}, \qquad \xi\ll L.$$

Neither result is universal at criticality or for long-range interactions, protected boundary modes, irregular shapes, or $\xi\sim L$.

Periodic boundaries remove physical surfaces, but they do not remove:

• momentum quantization;
• winding and wrapping processes;
• sector constraints;
• shell filling;
• critical infrared cutoffs;
• shape and aspect-ratio dependence.

Commensurability is equally important. A period-three structure fits a periodic chain only when the allowed momenta include $Q=2\pi/3$, which requires $L$ to be divisible by three. Comparing $L=12,15,18$ can reveal a clean subsequence while mixing $L=13,14,15$ can produce oscillations that imitate slow convergence.

Boundary Conditions on Lattices owns the construction of open, periodic, antiperiodic, and twisted Hamiltonians, their momentum grids, parity constraints, and boundary-specific diagnostics.

Shell Filling and Particle-Number Granularity

In a finite Fermi system, discrete one-particle modes fill in shells. A closed shell has all states in a degenerate group occupied; an open shell leaves a partially occupied multiplet. Consequently:

• the ground-state degeneracy can depend on $N$;
• the lowest particle-addition or particle-hole energy can jump at a shell closure;
• parity and boundary twists can change which modes lie at the Fermi surface;
• convergence with $N$ can oscillate rather than remain monotonic.

A concrete example is the periodic spinless tight-binding chain,

$$\epsilon(k) = -2t\cos k, \qquad k = \frac{2\pi m}{L},$$

at half filling. When $L=4r$, the grid contains $k=\pm\pi/2$, where $\epsilon(k)=0$, and the Fermi level is degenerate. When $L=4r+2$, those two momenta are absent. The nearest occupied and empty levels lie on opposite sides of $\pi/2$, giving the particle-hole spacing

$$\Delta_{\mathrm{ph}}(L) = 4t\sin\left(\frac{\pi}{L}\right) \simeq \frac{4\pi t}{L}.$$

The two congruence classes approach the same bulk band while displaying sharply different finite spectra. A boundary twist shifts the grid again.

The correct bulk sequence normally controls density or filling as the system grows. Even then, the nearest allowed momentum to the Fermi surface moves irregularly with size and shape. Twist averaging can smooth one-body shell sampling, but it does not erase interaction, topology, or sector effects.

Finite-$N$ ensemble choices are also physical. A canonical system has exactly fixed particle number; a grand-canonical description allows number fluctuations. Their local bulk predictions may become equivalent under suitable conditions, but at finite $N$ the fluctuations, shell occupation, and addition spectra can differ measurably.

The Ideal Fermi Gas owns detailed Fermi scales and shell cautions. The purpose here is the diagnosis: nonmonotonic size dependence may be structured shell physics, not failed convergence or random numerical noise.

Correlation-Length Cutoffs and Rounded Transitions

For a local bulk observable in a short-range system, the ratio $L/\xi$ often organizes finite-size behavior:

• if $L\gg\xi$ and the probe is far from a boundary, local behavior may already be bulk-like;
• if $L\lesssim\xi$, the system cannot contain the full correlation structure;
• if $\xi$ grows toward or beyond $L$, size itself becomes the infrared cutoff.

Near a continuous transition, a bulk correlation length may grow as

$$\xi(g) \sim \xi_0\lvert g-g_c\rvert^{-\nu}.$$

Here $\xi_0$ is a nonuniversal amplitude and $\nu>0$ is the correlation-length critical exponent for the assumed continuous transition.

The crossover $\xi(g)\sim L$ suggests a rounded critical window of scale

$$\lvert g-g_c\rvert \sim \left(\frac{\xi_0}{L}\right)^{1/\nu}$$

under the standard hypotheses. If the Hilbert space is finite dimensional and $H(g)$ depends analytically on the real coupling $g$, then the finite positive-temperature partition function is finite and analytic. Its peaks and rapid crossovers are not thermodynamic singularities. At zero temperature, exact symmetry-protected level crossings can still make a finite ground-state energy nonanalytic; the caveat does not turn one finite feature into proof of a bulk transition.

Finite size therefore rounds and shifts critical signatures rather than merely adding random noise. It can be informative: the way a susceptibility peak, dimensionless ratio, or gap changes with $L$ encodes long-distance physics.

Finite-size effects are the broad physical phenomenon. Finite-size scaling is a specialized asymptotic framework, especially powerful near critical points. A polynomial fit in $1/L$ is not automatically finite-size scaling.

Exponent dictionaries, universal scaling functions, correction terms, and data collapse belong to Critical Exponents and Scaling; fit windows, covariance, crossings, and robustness tests belong to Finite-Size Scaling in Numerics. First-order, Berezinskii–Kosterlitz–Thouless, disordered, anisotropic, and long-range cases require their own scaling logic.

Superfluidity in Condensed Matter applies these finite-size and BKT cautions to neutral films and material evidence; this page retains the general extrapolation contract.

Reflections, Revivals, and Recurrences

Finite space also creates finite time scales. In a short-range lattice system with an effective propagation speed $v$, a disturbance a distance $d_{\partial}$ from a boundary can return after a geometry-dependent time of order

$$t_{\mathrm{return}} \sim \frac{2d_{\partial}}{v}.$$

Let $t_{\mathrm{local}}$ denote the microscopic transient or local relaxation time after which the observable has entered the regime being studied. A useful local-dynamics window is then

$$t_{\mathrm{local}} \ll t_{\mathrm{obs}} \ll t_{\mathrm{return}},$$

when such a separation exists. Periodic systems replace reflection with wraparound. Long-range interactions, unbounded velocities, disorder, and small geometries require a different audit.

A finite-dimensional isolated system, or a state supported on finitely many discrete energy levels, produces quasiperiodic evolution. Revivals or recurrences can appear because the contributing phases eventually realign approximately or exactly. Finite spatial extent alone does not ensure this spectrum condition. Several time scales must remain distinct:

• a boundary-return or wraparound time;
• a dephasing or relaxation time for the chosen observable;
• a Heisenberg time $t_{\mathrm H}\sim2\pi\hbar/\delta$ estimated from the local mean level spacing $\delta$ in a declared energy window and symmetry sector;
• a Poincaré recurrence time for returning close to an initial state.

The last can be vastly longer than the others, and a Heisenberg-time estimate is not an exact recurrence theorem. Time-Dependent Correlations owns the detailed spectral and recurrence analysis.

Trapped and Inhomogeneous Systems

A cold-atom cloud is finite in several independent ways:

• particle number $N$ is finite and may fluctuate between shots;
• the trap produces discrete level spacings such as $\hbar\omega_i$;
• the density varies across the cloud;
• each cloud radius $R_i$ competes with correlation and healing lengths;
• temperature, imaging resolution, and observation time limit what is resolved.

There is no single substitution $V=L^d$ that turns every trapped cloud into a homogeneous box. One should compare $\hbar\omega_i$ with $k_{\mathrm B}T$, chemical-potential differences, interaction scales, and probe resolution; identify whether a local-density approximation is justified; and separate a finite trapped sequence from a homogeneous thermodynamic one.

An appropriate trap thermodynamic sequence may require the confinement to weaken as atom number grows. Keeping the trap fixed while adding particles defines a different family and can change density, occupations, interactions, and resolution scales rather than merely reducing a correction.

Quantum Gases in Traps owns the precise trap thermodynamic sequences, state counting, semiclassical limits, and local-density profiles. Experimental preparation, calibration, and imaging protocols belong to the relevant AMO platform pages.

Three Worked Finite-Size Audits

A gapless ring

A ring supports a branch $\epsilon(q)-\epsilon(0)=A\lvert q\rvert^z$. Before calling the lowest observed energy “the gap,” record:

1. the boundary phase and allowed momenta;
2. the symmetry sector of the ground and excited states;
3. whether $q=2\pi/L$ is actually allowed;
4. whether temperature or broadening exceeds $A(2\pi/L)^z$;
5. whether the same state identity is followed as $L$ changes.

Only after those checks does an $L^{-z}$ sequence support a quantization-gap interpretation.

A gapped open chain

Suppose a local observable differs from its bulk value only within a boundary layer of thickness $\xi$. A probe at the center can have exponentially small boundary contamination while the average over every site carries a boundary-to-volume correction. The statement “$L=20$ is large enough” can therefore be true for the central density and false for the total edge-sensitive response.

Changing to periodic boundaries tests this interpretation, but it also defines a different finite Hamiltonian and can remove the edge phenomenon one intended to study.

A small exact-diagonalization sequence

An eigensolver returns normalized residuals below $10^{-12}$ for periodic clusters of $L=12,15,18$ sites. A structure-factor peak at $Q=2\pi/3$ grows with $L$.

• Controlled numerical fact: each small residual certifies a nearby eigenvalue of the declared Hermitian matrix. Individual eigenvectors and structure factors are also stable only after the target level or invariant subspace is isolated and convergence is checked within any near-degenerate manifold.
• Favorable geometry: all three lengths are commensurate with period three.
• Unresolved physical question: the peak may reflect growing order, a long crossover, or the chosen commensurate subsequence.
• Needed bulk evidence: additional sizes or shapes, normalized observables, boundary and sector checks, and a justified scaling analysis.

The tiny residual controls numerical diagonalization error. It says nothing by itself about finite-size drift or the existence of long-range order.

A Reusable Finite-Size Ledger

Before interpreting a finite result, record:

1. Size variables: every length, volume, site count, particle number, graph diameter, cutoff, and observation time that changes.
2. Family: geometry, dimension, aspect ratio, boundary conditions, couplings, density or filling, and state preparation.
3. Observable: its support, normalization, quantum numbers, and distance from a boundary.
4. Physical scales: correlation, localization, screening, healing, or ordering lengths; mode spacings; gaps; temperature; and propagation speeds.
5. Gap identity: the two states, their sectors, and how they are tracked across sizes.
6. Resolution: numerical tolerance, spectral broadening, probe bandwidth, imaging scale, temperature, and Fourier window.
7. Mechanism: boundary, mode quantization, shell, parity, commensurability, critical rounding, edge hybridization, tunneling, recurrence, or trap inhomogeneity.
8. Pattern: monotonic, oscillatory, algebraic, exponential, or not yet asymptotic.
9. Alternative explanations: at least one competing mechanism consistent with the accessible sizes.
10. Claim strength: exact finite statement, controlled bound, scaling evidence, qualified bulk consistency, or established thermodynamic conclusion.

The ledger converts “large enough” into a checkable statement about one observable and one target accuracy.

Evidence Ladder for a Finite-Size Bulk Inference

Few-Body versus Many-Body Physics owns the classification of finite-system statements and thermodynamic claims. Once that claim type is explicit, this ladder grades the mechanism-specific finite-size evidence supporting the comparison:

1. Exact finite statement: “For this Hamiltonian and sector at $L=18$, the gap is …”
2. Mechanism diagnosis: “The leading variation is consistent with momentum quantization on this geometry.”
3. Controlled sequence: “The same normalized observable was evaluated over several sizes, boundaries, and relevant subsequences with smaller numerical error.”
4. Asymptotic evidence: “Competing correction forms and fit windows were tested, and the result is stable over a justified regime.”
5. Bulk conclusion: “The specified limit exists or is supported strongly enough to warrant the stated phase, gap, or critical claim.”

Do not jump from level 1 to level 5. A finite-system result can be valuable without being rewritten as a thermodynamic claim.

Canonical Boundaries

This page owns the cross-cutting physical taxonomy of finite-size mechanisms and the observable-and-resolution audit. Related canonical homes retain narrower subjects:

• Thermodynamic Limit owns limiting sequences, existence, boundary independence, singularities, and noncommuting limits.
• Extensive and Intensive Quantities owns bulk, surface, edge, subextensive, and superextensive scaling language.
• Boundary Conditions on Lattices owns detailed finite-Hamiltonian closures, twists, momentum grids, and parity constraints.
• Critical Exponents and Scaling owns universal critical scaling theory and correction terms.
• Finite-Size Scaling in Numerics owns regression, covariance, crossings, collapse, uncertainty, and robust extrapolation.
• Exact Diagonalization owns basis construction, eigensolver validation, and exactness within a finite representation.
• Time-Dependent Correlations owns detailed dephasing, discrete spectral sums, and recurrence theory.
• Quantum Gases in Traps owns trap thermodynamic limits, state counting, and local-density descriptions.

Common Pitfalls

“Periodic boundaries remove finite-size effects.” They remove physical surfaces, but mode quantization, wrapping paths, sector restrictions, shells, and critical infrared cutoffs remain. Name which mechanism the boundary choice suppresses.

“The gap closes with size, so the system is critical.” First identify both levels. Critical modes, Goldstone modes, symmetry partners, edge hybridization, and commensurability can all produce small gaps.

“The solver converged, so the bulk result converged.” Report numerical tolerance and physical size drift separately. Exact diagonalization is exact for its finite matrix, not for an infinite system.

“Larger is always closer.” Shell filling, parity, aspect ratio, and ordering-wavevector compatibility can make convergence oscillatory. Compare controlled subsequences and explain why they belong to the same target family.

“A smooth finite spectrum proves irreversible decay.” State the physical linewidth and observation window. Plotting broadening or unresolved lines does not create a lifetime.

“Boundary terms vanish, so edges do not matter.” Their fraction can vanish while an edge observable remains order one and scientifically central. Distinguish bulk averages from boundary-local measurements.

“One rounded peak is a phase transition.” Call it a finite-system crossover or pseudocritical marker until a declared limiting argument and scaling evidence support the stronger statement.

Exercises

Exercise 1: Quantization gap from a dispersion

A periodic one-dimensional system has a low-energy branch

$$\epsilon(q)-\epsilon(0) = A\lvert q\rvert^z,$$

and the first allowed nonzero momentum is $2\pi/L$. Derive the leading size dependence of the gap. Give the results for $z=1$ and $z=2$, and name two reasons the coefficient might change.

Solution

Substituting the first allowed momentum gives

$$\Delta_L = A\left(\frac{2\pi}{L}\right)^z.$$

Thus $\Delta_L=2\pi A/L$ for $z=1$ and $\Delta_L=4\pi^2A/L^2$ for $z=2$. A boundary twist can shift the allowed momenta, and a symmetry or conservation law can forbid the nominal first mode. Interactions can also renormalize the coefficient $A$ without changing the leading power in a stable scaling regime.

Exercise 2: An anisotropic boundary layer

A rectangular strip is periodic along $x$ with length $L_x$ and open along $y$ with width $L_y$. An observable differs by order one from its bulk value only within distance $\xi$ of either open edge, with $L_y\gg\xi$. Estimate the contaminated area fraction. What happens if (a) $L_x\to\infty$ at fixed $L_y$, or (b) both lengths grow at fixed aspect ratio?

Solution

The two boundary layers have total area approximately $2L_x\xi$, while the strip area is $L_xL_y$. Their fraction is therefore

$$f_{\partial} \simeq \frac{2\xi}{L_y}.$$

Taking $L_x\to\infty$ at fixed width does not reduce this fraction: the family remains a finite-width strip. If both lengths grow at fixed aspect ratio and $\xi$ remains finite, then $L_y$ grows and $f_{\partial}$ vanishes as an inverse linear size. Increasing the site count is insufficient unless every dimension relevant to the observable grows appropriately.

Exercise 3: Diagnose three small gaps

A sequence displays one of the following behaviors: (a) $\Delta_L\to0.8J$; (b) $\Delta_L\propto L^{-1}$; (c) $\Delta_L\propto e^{-L/\xi}$. Give plausible interpretations and explain why the size law alone does not identify the gap.

Solution

(a) is consistent with a nonzero bulk excitation gap. (b) is consistent with a linearly dispersing critical or Goldstone mode. Because $1/V$ also scales as $1/L$ in a one-dimensional geometry, the same power could describe an Anderson tower when a continuously ordered phase is actually admissible; ordinary short-range one-dimensional systems do not generically permit that interpretation. (c) is consistent with tunneling between symmetry-related states or hybridization of exponentially localized edge modes.

The interpretations are not unique. One must track symmetry sectors, momentum, boundary dependence, spatial profiles, overlaps, and matrix elements. The phrase “the gap” should be replaced by a precise state-to-state energy difference.

Exercise 4: A two-dimensional commensurability test

A periodic rectangle has allowed momenta $k_x=2\pi n_x/L_x$ and $k_y=2\pi n_y/L_y$. A candidate ordered state has wavevector $\mathbf Q=(\pi,2\pi/3)$. Which clusters among $(L_x,L_y)=(12,9),(15,9),(16,10),(18,12)$ contain $\mathbf Q$ exactly? Explain why a controlled comparison must record both components.

Solution

The $x$ component requires $n_x=L_x/2$, so $L_x$ must be even. The $y$ component requires $n_y=L_y/3$, so $L_y$ must be divisible by three. Thus $(12,9)$ and $(18,12)$ contain $\mathbf Q$ exactly; $(15,9)$ fails the $x$ condition, while $(16,10)$ fails the $y$ condition.

A cluster can therefore be large yet frustrate one component of the candidate pattern. Mixing compatible and incompatible rectangles into one smooth fit can disguise commensurability oscillations as an ordinary size correction. A controlled study should compare justified aspect ratios and compatibility subsequences, then test whether they approach the same observable limit.

Exercise 5: Quantization, broadening, and return time

Spectra from two isolated chains of lengths $L$ and $2L$ show three changes: the low-energy line spacing roughly halves, a late revival moves from $t_R$ to $2t_R$, and the plotted linewidth $\eta$ is unchanged. A report calls $\eta$ a decay rate and the smoother $2L$ spectrum a continuum. Classify the three observations and list the checks needed before accepting either interpretation.

Solution

The halved line spacing is consistent with finite-volume mode quantization if both calculations use the same branch, sector, and boundary convention. The delayed revival is consistent with a propagation or wraparound time proportional to system length. Neither observation by itself implies irreversible decay.

An unchanged plotting width $\eta$ is an imposed resolution scale unless it has been independently tied to a physical lifetime. A denser set of broadened finite-system lines can look smooth without becoming a continuum. One should vary $\eta$, compare it with experimental or numerical frequency resolution, repeat the calculation at more sizes, track corresponding states and symmetry sectors, and test the time signal before the first return. A decay-rate claim additionally needs a linewidth stable against those controls and a physical mechanism producing it.

Vortex Matter, Pinning, and Flux Flow applies this finite-size discipline to vortex-lattice order, melting or glass claims, barriers, observation times, and driven mixed-state response; finite samples and finite records do not by themselves establish a thermodynamic vortex phase.

References

• M. N. Barber, “Finite-Size Scaling,” in C. Domb and J. L. Lebowitz, eds., Phase Transitions and Critical Phenomena, Vol. 8, Academic Press (1983), pp. 145–266.
• M. Brack and R. K. Bhaduri, Semiclassical Physics, 2nd ed., Westview Press (2003).
• M. Campostrini, A. Pelissetto, and E. Vicari, “Finite-Size Scaling at Quantum Transitions”, Physical Review B 89, 094516 (2014).
• J. L. Cardy, ed., Finite-Size Scaling, North-Holland (1988).
• M. E. Fisher and M. N. Barber, “Scaling Theory for Finite-Size Effects in the Critical Region”, Physical Review Letters 28, 1516–1519 (1972).
• W. Ketterle and N. J. van Druten, “Bose–Einstein Condensation of a Finite Number of Particles Trapped in One or Three Dimensions”, Physical Review A 54, 656–660 (1996).
• V. Privman, ed., Finite Size Scaling and Numerical Simulation of Statistical Systems, World Scientific (1990).
• A. W. Sandvik, “Computational Studies of Quantum Spin Systems”, AIP Conference Proceedings 1297, 135–338 (2010).