Why the size of the box matters

A finite box can introduce an energy scale even when the infinite-volume theory is massless. A positive level spacing at one box size therefore does not establish a mass gap in unbounded space.

Prerequisite: the lattice definition. The example below is a free scalar field, used to isolate the role of volume rather than model Yang–Mills dynamics.

By the end:

• Exhibit a box-induced gap that vanishes as the box grows.
• State which physical quantities must be controlled uniformly in volume.

An example without a hidden zero mode

Put a real scalar field of mass $m$ in a spatial cube $[0,L]^3$ and impose Dirichlet boundary conditions: the field vanishes on the boundary. Keep time unbounded. The normal modes are products of sine functions,

$$f_{\mathbf n}(\mathbf x)\propto \prod_{i=1}^3\sin\frac{\pi n_i x_i}{L}, \qquad n_i=1,2,\ldots.$$

Their frequencies follow from the Klein–Gordon equation:

$$\omega_{\mathbf n} =\sqrt{m^2+\frac{\pi^2}{L^2} (n_1^2+n_2^2+n_3^2)}.$$

After subtracting the vacuum energy, the free Hamiltonian is a sum of oscillator number operators. The first excitation has energy

$$\Delta_L=\sqrt{m^2+\frac{3\pi^2}{L^2}}.$$

If $m=0$, every finite box has $\Delta_L>0$, but

$$\Delta_L=\frac{\sqrt3\pi}{L}\longrightarrow0 \quad\text{as }L\to\infty.$$

If $m>0$, the same sequence tends to $m$. The behavior across volumes distinguishes a mass from a scale induced by the boundary.

Dirichlet conditions are part of the example. With periodic conditions a massless scalar has a spatial zero mode that needs separate treatment. Silently deleting it would change the model and conceal an important assumption.

Uniformity is the central question

Suppose a family of theories has Hamiltonians $H_L$ and normalized vacua $\Omega_L$. A useful finite-volume statement might be

$$\langle\psi,H_L\psi\rangle \geq \Delta_*\|\psi\|^2, \qquad \psi\perp\Omega_L,$$

for a common $\Delta_*>0$ and every sufficiently large $L$ in the appropriate domains.

The word “common” does the work. A bound with $\Delta_*=\Delta_*(L)$ can become useless as the box grows. Even a uniform bound still requires an argument connecting the finite-volume theories to the limiting state space and Hamiltonian.

The uniform-estimates bridge explains this pattern with simpler limits. In a constructive proof, one tracks the parameters on which every constant depends.

Finite volume can change more than the lowest energy

Multiparticle levels become discrete in a box. Their positions depend on momenta, interactions, and boundary conditions. A state that is unstable in infinite volume can appear among a collection of discrete finite-volume levels. Interpreting those levels requires more information than locating an exponential in one correlator.

Boundary conditions can also influence flux sectors and extended observables. In gauge theory, specify whether directions are periodic, open, or twisted. A result for one choice should not silently become a theorem for all choices.

Numerical spectroscopy therefore compares several physical volumes at comparable lattice spacings. It also investigates whether the chosen operator basis can distinguish the states of interest. The SU(N) spectrum study by Athenodorou and Teper is a concrete reference for these practical issues.

The Euclidean time direction has a separate role

A finite periodic Euclidean time extent $\mathcal T$ usually describes a thermal trace:

$$C_{\mathcal T}(t)= \frac{\operatorname{Tr}\!\left( e^{-(\mathcal T-t)H}O^\dagger e^{-tH}O \right)} {\operatorname{Tr}(e^{-\mathcal T H})}.$$

It approaches a vacuum expectation only under appropriate large-$\mathcal T$ control. For a dominant state of energy $E$ and sufficiently suppressed thermal contributions, one often obtains the approximate form

$$C_{\mathcal T}(t)\simeq A\left(e^{-Et}+e^{-E(\mathcal T-t)}\right).$$

The second term propagates around the periodic time direction. Near the midpoint, a simple logarithmic ratio can therefore depart from the infinite-time effective mass even for perfect data.

Spatial volume, time extent, and lattice spacing are distinct controls. Changing one does not automatically resolve the others.

What to ask when reading a result

Write down the physical spatial length, the Euclidean time extent, the boundary conditions, and the parameters held fixed. Then ask whether the claimed estimate remains informative as each necessary limit is taken.

For an area-law or correlation theorem, inspect whether constants depend on the box, the boundary data, or the cutoff. Chatterjee’s lattice confinement criterion shows why boundary-uniform correlation assumptions can be essential rather than technical decoration.

Exercises

1. Two masses in the same box

Compare $\Delta_L$ for $m=0$ and $m>0$. What distinguishes their large-$L$ limits?

Hint. Expand the square root when $mL$ is large.

Solution. The massless gap tends to zero as $1/L$. For $m>0$, $\Delta_L=m+3\pi^2/(2mL^2)+O(L^{-4})$ in this free Dirichlet example. The finite positive limit is the physical mass.

2. A misleading uniform statement

Someone proves $\Delta_L\geq 1/(L+1)$ for every box. Is that a volume-uniform positive lower bound?

Hint. Take the infimum over $L$.

Solution. No. The bound is valid for all boxes, but its right-hand side tends to zero. “For all volumes” does not mean “with a positive constant independent of volume.”

Continue at your own depth

A gap made by the box: interactive lab uses a one-dimensional Dirichlet interval, so its massless first frequency is $\pi/L$ rather than the three-dimensional cube value $\sqrt{3}\pi/L$ above. The boundary-induced gap disappears in either example as the physical box grows.

Read further

• Athenodorou and Teper, lattice spectra and physical limits.
• Chatterjee, A probabilistic mechanism for quark confinement.

Previous: A lattice definition · Course overview · Next: The continuum limit