Topology and Lattice Index Diagnostics

Continuum gauge fields separate into topological sectors under smoothness and boundary assumptions. At nonzero lattice spacing, generic links do not carry a unique integer topological charge: local discretizations, geometric definitions, and fermionic indices agree only within stated smoothness regimes and toward a controlled continuum limit. Reliable topology measurements must distinguish ultraviolet dislocations, definition dependence, slow sector tunneling, and finite-volume sampling.

Required background. The Wilson gauge action supplies the regulated ensemble; large gauge transformations and topological sectors supply the continuum classification.

Helpful background. Review topological susceptibility and the Dirac index and zero modes. Gradient flow gives one controlled smoothing framework.

Continuum target and lattice definitions

Local convention and regulator card. Work in four-dimensional Euclidean space with Hermitian generators, $\epsilon_{0123}=+1$ , unrestricted repeated Lorentz-index sums, and $\widetilde F_{\mu\nu}^a=\frac12\epsilon_{\mu\nu\rho\sigma}F_{\rho\sigma}^a$ . The continuum charge and the overlap index are oriented so that $Q=+1$ for a self-dual unit instanton and $Q_{\mathrm{index}}=n_+-n_-$ . Every lattice result must state its charge definition, admissibility or flow condition, boundary condition, volume, and sector-sampling history.

For smooth four-dimensional $SU(N)$ fields with the chapter’s Hermitian generators,

Q=\frac{g^2}{32\pi^2}\int d^4x\, F_{\mu\nu}^a\widetilde F_{\mu\nu}^a =\frac{g^2}{64\pi^2}\int d^4x\, \epsilon_{\mu\nu\rho\sigma}F_{\mu\nu}^aF_{\rho\sigma}^a

is integer under suitable finite-action boundary conditions. The factor $1/(64\pi^2)$ in the epsilon form is required because both index pairs are summed and $\widetilde F$ already contains $1/2$ . As a unit check, a self-dual $SU(2)$ instanton obeys $\int d^4x\,F_{\mu\nu}^aF_{\mu\nu}^a=32\pi^2/g^2$ , so the first expression gives $Q=+1$ ; an anti-self-dual instanton gives $-1$ .

A field-theoretic lattice estimator replaces $F_{\mu\nu}$ by a clover or improved loop combination,

Q_L=a^4\sum_x q_L(x).

On a rough configuration, $Q_L$ need not be integer and may require multiplicative renormalization and additive contact-term subtraction in $\langle Q_L^2\rangle$ . The continuum-looking formula does not by itself confer topology.

A geometric construction can assign an integer by interpolating links within cells, but only after branch and admissibility conditions are fixed. Different constructions can disagree on dislocations whose size is comparable with $a$ .

The smoothness domain needed for a geometric integer and its stability is constructed in Lüscher 1982, pp. 39–48.

Fermionic index diagnostics

If a $\gamma_5$ -Hermitian lattice Dirac operator satisfies the Ginsparg–Wilson relation

\gamma_5D+D\gamma_5=aD\gamma_5D,

then the index can be written

Q_{\mathrm{index}}=n_+ - n_- =\operatorname{Tr}\!\left[\gamma_5\left(1-\frac{aD}{2}\right)\right]

in this volume’s sign convention. Here $n_\pm$ count exact zero modes with $\gamma_5=\pm1$ . Authors who define the index as $n_- - n_+$ must reverse both this trace formula and the associated topological-charge sign. The index is an exact integer for a suitable finite operator, but its stability under continuous changes of the gauge field requires locality and a smoothness or admissibility domain. A near-zero eigenvalue of an approximate chiral operator is not automatically an exact index mode.

An explicit overlap operator with exact lattice chirality and zero-mode structure is given in Neuberger 1998, pp. 141–144.

The strongest check compares a fermionic index, a geometric definition, and a suitably improved field-theoretic charge on increasingly fine ensembles. Agreement after controlled smoothing is evidence for a shared continuum observable; disagreement identifies a cutoff or sampling problem.

Susceptibility and sector sampling

At vacuum angle zero,

\chi_t=\frac{\langle Q^2\rangle-\langle Q\rangle^2}{V_4}.

Because $Q$ is global and often evolves slowly as $a$ decreases, the effective number of independent sectors can be far smaller than the number of stored configurations. Monitor the history of $Q$ , its integrated autocorrelation time, transition counts, and results in subvolumes or at open temporal boundaries where appropriate.

A chain frozen in one sector can produce apparently precise local observables. If fixed-topology corrections are used, their expansion assumes large $\chi_tV_4$ and knowledge of sector dependence; it cannot manufacture missing information when the volume is too small.

Resampling must respect the Markov chain. Blocks shorter than the topological autocorrelation time underestimate uncertainty, while a histogram with several sectors does not establish equilibrium weights unless repeated starts or independent streams agree.

Smoothing without changing the question

Cooling, stout smearing, and gradient flow suppress ultraviolet fluctuations. They can reveal integer-like plateaus, but the smoothing prescription introduces a scale. A controlled study records a physical smoothing radius, checks that it is well separated from both $a$ and $L$ , and takes $a\to0$ at fixed physical flow time before any optional $t\to0$ interpretation.

An admissible continuous flow cannot change an integer topological sector without crossing a rough configuration where the definition becomes ambiguous. Thus a sudden change in $Q$ while every plaquette remains inside the stated admissibility bound is an implementation or orientation error.

A finite-lattice diagnostic protocol

Question	Minimum evidence	Failure mode isolated
Is the charge definition valid?	Formula, orientation, normalization, smoothness domain	Noninteger rough-field estimator called topological
Do definitions agree?	Geometric, field, or index comparison versus $a$	Dislocations or normalization error
Are sectors sampled?	Histories, transitions, $\tau_{\mathrm{int},Q}$ , streams	Topological freezing
Is susceptibility controlled?	Connected variance, volume sequence, contact treatment	Finite-volume or ultraviolet bias
Is smoothing controlled?	Fixed physical $t$ , discretization comparison, flow window	Arbitrary smoothing dependence
Is the continuum claim justified?	Agreement and scaling along constant physics	One-spacing integer plateau overinterpreted

The regime map deliberately keeps topological and gradient-flow checks connected but distinct: flow can define renormalized observables and improve a charge estimator, yet correct sector sampling remains an ensemble property.

Gauge configurations branch into gauge-invariant loops, gauge-fixed correlators, strong-coupling series, topology, and gradient-flow observables, each with a distinct validity test. — Topology requires agreement among definitions plus verified sector sampling and volume control. Positive flow can assist the definition test, but it cannot supply missing tunneling. The diagram is schematic and not to scale.

Adversarial failure: a precise susceptibility from a frozen sector

Consider $10^5$ stored configurations whose flowed charge is stably integer-valued but remains $Q=1$ throughout. Two improved field definitions and the overlap index can agree on every configuration, and bootstrap errors can become arbitrarily small. Nevertheless the connected estimator gives

\widehat\chi_t= \frac{\overline{Q^2}-\overline Q^{\,2}}{V_4}=0,

because the chain never sampled the equilibrium sector weights. Agreement among charge definitions tests the observable on sampled fields; it does not test ergodicity. Independent streams, transitions, autocorrelation analysis, or a controlled fixed-topology treatment are therefore indispensable.

Observable-level validation checklist

On a smooth oriented test field, verify the epsilon normalization, $Q=+1$ for a self-dual unit instanton, and the corresponding overlap-index sign.
Measure the Ginsparg–Wilson residual, isolate exact chiral zero modes, and confirm that $\operatorname{Tr}[\gamma_5(1-aD/2)]=n_+-n_-$ .
Compare field-theoretic, geometric, and fermionic charges versus spacing at fixed physical smoothing radius rather than merely rounding one estimator.
Inspect charge histories, sector transitions, independent streams, and $\tau_{\mathrm{int},Q}$ before estimating $\chi_t$ .
Repeat the susceptibility at another four-volume and smoothing window, including contact-term or boundary corrections appropriate to the definition.

Common pitfalls

Rounding $Q_L$ to the nearest integer. Rounding hides definition dependence and biases $\chi_t$ . Establish a plateau and continuum agreement before interpreting clusters near integers.

Using the plaquette autocorrelation time. Topological modes can be orders of magnitude slower. Diagnose $Q$ itself and observables coupled to it.

Letting flow time drift in physical units. Holding $t/a^2$ fixed sends the physical smoothing radius to zero as $a\to0$ . Decide which limit is intended and state it.

Learning outcomes

Given a lattice gauge field and charge prescription, translate its orientation and normalization, compute the fermionic index sign, and test agreement with an independent definition on smooth fields.
Given a topological-charge time series, distinguish dislocations, smoothing dependence, sector freezing, and finite-volume bias and determine whether a susceptibility estimate is supportable.

Exercises

A run remains in $Q=1$ throughout. What value does the naive connected estimator of $\chi_t$ give, and why is it unreliable?

Solution

It gives $(\langle Q^2\rangle-\langle Q\rangle^2)/V=0$ . This reflects absence of sector fluctuations in the chain, not the equilibrium susceptibility.

Show that the Ginsparg–Wilson trace receives contributions only from special real modes and reduces to the chiral zero-mode difference.