QFT Bridge: Instantons and Saddle Points

Saddle-point methods organize functional integrals around stationary configurations. Quantum-mechanical instantons provide the cleanest nontrivial example: a finite-action Euclidean path contributes an exponentially small sector that perturbation theory around one minimum cannot see.

Quantum field theory keeps that logic but changes almost every technical ingredient. The integration variable is a field, the Hessian is a differential operator on spacetime, gauge redundancy may be present, determinants require ultraviolet renormalization, and finite-action boundary conditions can carry topology.

This page is a comparison and audit guide. Instantons in Quantum Mechanics owns the one-dimensional saddle construction. Fluctuation Determinants Preview owns the quantum-mechanical prefactor logic. Bridge to QFT Instantons owns the canonical path-to-field translation. The purpose here is to classify the ingredients of a QFT saddle argument and identify what must be checked before trusting its conclusion.

Canonical Boundaries

Question	Canonical page	Role of this page
How is a quantum-mechanical instanton constructed?	Instantons in Quantum Mechanics	use its result as the $0+1$ dimensional prototype
How does Euclidean time encode tunneling boundary data?	Euclidean Time and Imaginary-Time Action	check signs, vacuum subtraction, and finite action
How does a determinant prefactor arise?	Fluctuation Determinants Preview	distinguish ordinary, zero, and negative modes
How do paths become field configurations?	Bridge to QFT Instantons	supply the canonical conceptual bridge
How should a saddle argument in QFT be audited?	this page	classify contour, boundary conditions, modes, measure, renormalization, topology, and observable

The same formula may appear briefly in several rows of this map, but its derivation has one canonical home.

Universal Saddle Pattern

Consider a Euclidean functional integral written schematically as

$$Z_E = \int_{\mathcal C} \mathcal D\Phi\, \exp\left( -\frac{1}{\hbar}S_E[\Phi] \right).$$

The symbol $\mathcal C$ matters. It represents the integration cycle together with boundary conditions, reality conditions, gauge treatment, and any restriction to a sector. A formal solution of the field equation is not automatically a contributing saddle of this integral.

A stationary configuration $\Phi_\star$ satisfies

$$\left. \frac{\delta S_E}{\delta\Phi} \right|_{\Phi_\star} =0.$$

Writing

$$\Phi = \Phi_\star+\eta,$$

gives the local expansion

$$\begin{aligned} S_E[\Phi] &= S_E[\Phi_\star] \\ &\quad+ \frac12 \langle\eta, \mathcal M_\star\eta\rangle + O(\eta^3), \end{aligned}$$

where $\mathcal M_\star$ is the Hessian or quadratic fluctuation operator. After zero modes, negative modes, gauge directions, and regularization have been handled, one saddle sector has the schematic form

$$\begin{aligned} Z_\star &\sim e^{-S_E[\Phi_\star]/\hbar} \\ &\quad\times J_{\mathrm{coll}} \\ &\quad\times \left( \det{}'\mathcal M_\star \right)^{-1/2} \\ &\quad\times \bigl[1+O(\hbar)\bigr]. \end{aligned}$$

This is an organizational formula, not a universal normalized answer. $J_{\mathrm{coll}}$ denotes collective-coordinate Jacobians and measures; the prime excludes modes requiring separate treatment. Gauge fixing, ghost factors, reference determinants, counterterms, and operator insertions are suppressed.

Before Calling a Configuration a Saddle

Check the following in order.

Integration cycle

State whether the calculation is Lorentzian, Euclidean, or defined by a complex deformation. A Euclidean stationary point need not lie on the original Lorentzian contour, and a formal Wick rotation requires analytic and boundary-condition assumptions. From Euclidean Time to Euclidean QFT gives the broader continuation map.

Boundary conditions

Specify the asymptotic field values, endpoints, periodicity, operator insertions, and sector. The boundary data determine which stationary configurations are relevant and whether the action is finite.

Finite action

Evaluate the action with the same vacuum subtraction and normalization used by the functional integral. A solution with divergent action does not contribute as an isolated finite-action semiclassical event in the usual way.

Stationarity

Verify the full variational equation, including surface terms. Solving the bulk Euler–Lagrange equation is insufficient when the boundary variation does not vanish.

Fluctuation spectrum

Classify positive, zero, negative, and gauge directions of the Hessian. The name of the classical solution does not determine this spectrum by itself.

Observable

State whether the saddle contributes to an energy splitting, a decay rate, a transition amplitude, a correlation function, a vacuum energy, or another quantity. The same saddle action can enter different observables with different prefactors and combinatorics.

Saddle Taxonomy

Terminology varies across subfields, so boundary conditions and mode content are more reliable than names.

Configuration	Defining feature	Typical role	Essential warning
vacuum saddle	stationary configuration in the reference sector	ordinary perturbation theory	does not include other saddle sectors
quantum-mechanical instanton	finite-action Euclidean path between degenerate minima	tunneling amplitude or level splitting	not a hidden real-time trajectory
anti-instanton	oppositely oriented instanton sector	sector sums and interference	orientation and topological charge conventions matter
bounce	Euclidean round trip associated with metastability	imaginary part and decay rate	normally has a negative mode
field-theory instanton	finite-action Euclidean field configuration, often between sectors	nonperturbative amplitudes or effective interactions	gauge, topology, fermion modes, and renormalization may enter
thermal saddle	periodic Euclidean configuration on a thermal circle	finite-temperature transition or rate	periodicity and thermal crossover change the problem
complex saddle	stationary point after complexifying fields or contours	steepest-descent decomposition	contribution depends on the integration cycle

A static soliton in Lorentzian spacetime is not automatically an instanton. A renormalon singularity is not, in general, a classical finite-action saddle. A bounce is not interchangeable with a degenerate-vacuum instanton.

From One Coordinate to a Field

For one quantum-mechanical coordinate,

$$S_E[x] = \int d\tau \left[ \frac{m}{2}\dot x^2 +V(x) \right].$$

A scalar field in $d$ spatial dimensions has the schematic action

$$S_E[\phi] = \int d\tau\,d^d x \left[ \frac12 (\partial_\mu\phi)^2 +V(\phi) \right].$$

The ordinary differential equation becomes a partial differential equation,

$$-\partial_\mu\partial_\mu\phi + V'(\phi) =0,$$

subject to finite-action boundary data. The increase in dimension changes the solution space, the asymptotic analysis, and the fluctuation spectrum.

Ingredient	Quantum mechanics	Quantum field theory
integration variable	path $x(\tau)$	field configuration $\Phi(\tau,\mathbf x)$
saddle equation	ordinary differential equation	partial differential or gauge-field equation
boundary	endpoints or limits in Euclidean time	Euclidean spacetime infinity, thermal circle, defects, or operator data
Hessian	one-dimensional differential operator	spacetime differential operator with internal indices
collective coordinates	often instanton center	position, size, orientation, gauge orientation, and other moduli
redundancy	usually absent in elementary models	gauge fixing and ghosts may be required
determinant	regulated spectral ratio	ultraviolet-divergent functional determinant requiring renormalization
topology	disconnected path sectors may suffice	homotopy or bundle data can label field sectors
output	splitting, amplitude, or decay width	correlator, effective vertex, vacuum amplitude, rate, or selection rule

The first column is a prototype, not a formula generator for the second.

Hessian Modes

Suppose

$$\mathcal M_\star\eta_n = \lambda_n\eta_n.$$

The sign and origin of $\lambda_n$ determine how the mode is treated.

Mode	Meaning	Treatment
positive	stable Gaussian direction	include in the regulated determinant
exact zero	continuous family of saddles	remove from the determinant and integrate over a collective coordinate
negative	unstable steepest-descent direction	define the contour carefully; may produce an imaginary part
gauge	redundant description, not a physical modulus	impose gauge fixing and include the associated Jacobian or ghost determinant
near-zero	weakly lifted modulus or interacting saddle sector	treat beyond a naive Gaussian approximation

Translational zero modes

If translating a quantum-mechanical instanton center does not change the action, then

$$\eta_0(\tau) \propto \frac{d x_\star}{d\tau}$$

is a zero mode. In a translationally invariant field theory, derivatives $\partial_\mu\Phi_\star$ generate the corresponding position modes when they are normalizable.

The zero eigenvalue is not inserted into an ordinary determinant. It is replaced by integration over the saddle position with a Jacobian fixed by the mode normalization.

Negative modes

A negative eigenvalue means the Euclidean action decreases along one local direction. For a metastable bounce, the negative mode is tied to analytic continuation and the imaginary part from which a decay rate is extracted. The existence, number, and contour treatment of negative modes are therefore part of the physical argument, not a removable numerical inconvenience.

Bounce Solutions owns the quantum-mechanical bounce construction and its connection to false-vacuum decay.

Gauge directions

An infinitesimal gauge transformation can look like a zero direction of the unfixed Hessian. Integrating it as an ordinary collective coordinate overcounts physically equivalent configurations. Gauge fixing separates redundancy from genuine moduli; the resulting Jacobian and ghost operators belong to the field-theory measure.

Determinants and Renormalization

In a one-dimensional quantum-mechanical problem, a prefactor can often be expressed through a regulated ratio such as

$$\left[ \frac{ \det\mathcal M_{\mathrm{ref}} }{ \det{}'\mathcal M_\star } \right]^{1/2}.$$

Even there, boundary conditions, zero-mode normalization, and the reference operator matter.

In QFT, the spectrum contains arbitrarily short-wavelength modes. A formal product of eigenvalues is ultraviolet divergent. A valid one-loop result must specify:

• the regularization method;
• gauge fixing and ghost contributions when relevant;
• the reference background or normalization;
• counterterms and renormalized parameters;
• the renormalization scale and scheme;
• any remaining infrared or volume dependence.

The exponent can also involve running couplings evaluated at a scale set by a collective coordinate. Thus the division between “classical exponent” and “prefactor” can depend on how the renormalized semiclassical expansion is organized.

Do not transfer a finite quantum-mechanical determinant ratio to a field theory by replacing $x$ with $\phi$.

Where Topology Enters

Finite action often requires a field to approach a vacuum configuration at Euclidean infinity. Identifying points at infinity can turn the boundary into a compact space, and the asymptotic field or gauge transformation may define a map into the vacuum manifold or gauge group. Homotopy data can then label disconnected sectors.

This is the structural origin of topological charge in many instanton problems. It is not automatic:

• a Euclidean saddle need not carry nonzero topological charge;
• a bounce can control decay without being a topological instanton;
• the same topological sector can contain several saddles;
• topology labels sectors but does not determine the full determinant or observable;
• gauge conventions and boundary conditions must be specified before writing a charge.

The explicit Yang–Mills preview and its action bound live in Bridge to QFT Instantons. The geometric background for homotopy and characteristic classes lives in the Geometry and Topology toolkit.

Observable First

The saddle is an ingredient of an observable, not the observable by itself.

Desired result	What the saddle supplies	What still has to be done
double-well splitting	one-event action and fluctuation data	sum instanton and anti-instanton sectors, normalize the kernel, extract energies
metastable decay rate	bounce action and mode spectrum	treat the negative mode, normalize time or volume, define the survival law
correlation function	saddle background and inserted fields	include operator insertions, collective-coordinate integrals, and normalization by $Z$
effective interaction	zero-mode and symmetry structure	saturate required modes, match operator normalization, run coefficients if needed
vacuum amplitude	saddle-sector weight	sum sectors, handle volume factors, compare competing saddles
topological response	sector and charge data	specify coupling to the topological term and the observable derivative

An exponential factor such as $e^{-S_\star/\hbar}$ supports only an exponent-level claim until the remaining steps are supplied.

Evidence Ladder

Keep five levels of claim separate.

1. Existence of a saddle

The field equation and boundary conditions admit a finite-action stationary configuration.

2. Classical exponential

The saddle action gives the leading exponential suppression in a stated semiclassical regime.

3. One-loop saddle contribution

The Hessian, zero modes, negative modes, gauge factors, and determinant regularization have been treated consistently.

4. Normalized observable

Sector sums, insertions, state normalization, time or volume factors, and analytic continuation have been assembled into a physical quantity.

5. Dominance and accuracy

Competing saddles, multi-saddle interactions, higher loops, infrared effects, and other nonperturbative mechanisms are known to be smaller in the stated regime.

Reaching one level does not imply the next. In particular, finding a beautiful classical solution does not establish that it dominates a measurable effect.

Saddle Audit

When reading or writing a QFT saddle argument, record:

observable and functional integral
integration cycle and continuation
boundary conditions and sector
finite-action saddle equation
classical action and control parameter
positive, zero, negative, and gauge modes
collective-coordinate measure
determinant regulator and renormalization scheme
operator insertions and selection rules
multi-saddle and competing-sector corrections
normalization, dimensions, and limiting checks
scope of the final claim

If any line is absent, narrow the claim to the level actually supported.

Relation to Perturbation Theory

A contribution of the form

$$\exp\left( -\frac{S_\star}{\hbar} \right)$$

is nonanalytic at $\hbar=0$ and is invisible at every finite order of an ordinary power series around a different saddle. This does not make perturbation theory irrelevant. Each saddle has its own fluctuation expansion, and the full answer may require perturbative and nonperturbative sectors together.

Multi-saddle sectors, large-order behavior, and ambiguity cancellation lead toward resurgence. Resurgence Preview introduces that connection without treating it as a universal solved structure in arbitrary QFTs.

What Transfers and What Must Be Rebuilt

Transfers:

• stationary action organizes a semiclassical sector;
• finite action and boundary data select relevant configurations;
• the classical action controls a leading exponential;
• quadratic fluctuations produce determinant structure;
• exact symmetries produce zero modes and collective coordinates;
• different saddles can encode effects absent from one local perturbative series.

Must be rebuilt in QFT:

• the functional measure and normalization;
• gauge fixing and ghost factors;
• ultraviolet regularization and renormalization;
• fermionic zero-mode treatment and selection rules;
• spacetime and internal collective coordinates;
• topological-sector definitions;
• the relation between a saddle sector and the chosen observable;
• dominance over competing nonperturbative mechanisms.

Common Mistakes

• Calling every stationary Euclidean configuration an instanton.
• Solving the bulk field equation without checking surface terms or finite action.
• Omitting the integration cycle when complex or negative modes are present.
• Treating a bounce as a degenerate-vacuum instanton.
• Including an exact zero eigenvalue in an ordinary Gaussian determinant.
• Treating gauge redundancy as a physical collective coordinate.
• Copying a quantum-mechanical determinant prefactor into QFT.
• Quoting a bare coupling in an instanton exponent without a scale or scheme.
• Assuming topology fixes the complete semiclassical contribution.
• Interpreting the saddle exponential as a normalized probability or rate.
• Claiming instanton dominance without comparing other saddles or mechanisms.
• Presenting a resurgence analogy as a theorem for an arbitrary field theory.

References

• S. Coleman, Aspects of Symmetry, Cambridge University Press, 1985.
• S. Coleman, “Fate of the False Vacuum: Semiclassical Theory,” Physical Review D 15, 2929–2936, 1977; erratum 16, 1248, 1977.
• C. G. Callan Jr. and S. Coleman, “Fate of the False Vacuum. II. First Quantum Corrections,” Physical Review D 16, 1762–1768, 1977.
• R. Rajaraman, Solitons and Instantons, North-Holland, 1982.
• G. ‘t Hooft, “Computation of the quantum effects due to a four-dimensional pseudoparticle,” Physical Review D 14, 3432–3450, 1976.
• A. M. Polyakov, Gauge Fields and Strings, Harwood Academic Publishers, 1987.
• J. Zinn-Justin, Path Integrals in Quantum Mechanics, Oxford University Press, 2005.

Exercises

Exercise 1: Instanton or bounce?

A Euclidean solution begins and ends at the same metastable minimum and has one negative fluctuation mode. Should it be treated as a degenerate-vacuum instanton that directly gives a level splitting?

Solution

No. Those features identify the bounce problem associated with metastable decay. The negative mode and analytic continuation are part of extracting an imaginary energy or decay rate. A degenerate-vacuum instanton instead connects distinct minima or sectors and contributes to transition amplitudes and level splitting after the relevant saddle sum. Use Bounce Solutions for the canonical construction.

Exercise 2: A translational family

Suppose $\Phi_\star(x-a)$ is a family of saddles with the same action for every position $a$. Show why differentiation with respect to $a$ produces a Hessian zero mode.

Solution

The saddle equation can be written as

$$\mathcal E[\Phi_\star(x-a)] =0.$$

Differentiate with respect to $a$:

$$\left. \frac{\delta\mathcal E}{\delta\Phi} \right|_{\Phi_\star} \frac{\partial\Phi_\star}{\partial a} =0.$$

The linearized equation operator is the Hessian $\mathcal M_\star$, up to the conventions used for fields and inner products. Therefore

$$\mathcal M_\star \frac{\partial\Phi_\star}{\partial a} =0.$$

The mode is tangent to the moduli family. It must be replaced by integration over $a$ with the appropriate Jacobian.

Exercise 3: Gauge zero direction

Why is an infinitesimal gauge transformation of a saddle not automatically integrated as an ordinary physical modulus?

Solution

Gauge-related fields represent the same physical configuration. Integrating freely along that direction would overcount the gauge orbit. One first fixes the gauge and includes the associated Jacobian or ghost determinant. Only residual parameters that label physically inequivalent solutions become genuine collective coordinates.

Exercise 4: Exponent versus rate

A calculation finds a finite-action bounce with action $B$ and reports $\Gamma=e^{-B/\hbar}$. What is missing from this equality?

Solution

The exponential supplies the leading suppression, not the complete normalized rate. The calculation still needs the determinant prefactor, zero-mode Jacobian, negative-mode contour, normalization by time and possibly volume, the relation to the metastable survival amplitude, dimensions, and corrections from other saddle sectors. A defensible intermediate statement is $\Gamma\propto e^{-B/\hbar}$ at exponent level in a specified semiclassical regime.

Exercise 5: Does finite action imply topology?

A scalar field has a localized finite-action Euclidean saddle. Does finite action alone prove that the saddle carries a nonzero topological charge?

Solution

No. Finite action constrains asymptotic behavior, but a nonzero topological charge requires a defined topological invariant and boundary map into the relevant vacuum manifold or gauge group. A bounce can have finite action without being a topological instanton. The theory, dimension, boundary conditions, and configuration sector must all be specified.