Feynman Rules

Feynman rules are a dictionary that translates each term in the perturbative expansion of a quantum field theory's correlation functions and scattering amplitudes into a set of diagrams and an associated algebraic expression. Key result: every scattering amplitude can be computed by drawing all topologically distinct diagrams at a given order, assigning propagators to internal lines, vertices to interactions, and external wavefunctions to incoming/outgoing particles, then integrating over undetermined loop momenta. Applies to all perturbative QFTs — QED, QCD, the electroweak theory, scalar models, and effective field theories. Origin: Richard Feynman (1948–1949), with Freeman Dyson supplying the systematic derivation from the Dyson series.

The Feynman rules are the operational core of perturbative quantum field theory: a finite recipe converting the abstract Dyson series — a time-ordered exponential of the interaction Hamiltonian — into concrete, calculable integrals. Each diagram is a mnemonic for a term in Wick's expansion of a vacuum expectation value of time-ordered fields, with lines representing two-point contractions (propagators) and vertices representing the interaction terms. They matter because they reduce the computation of physical observables — cross sections, decay rates, anomalous moments — to a bookkeeping problem that is at once intuitive and rigorously defined. The surprising fact is that diagrams which look like real spacetime processes are in fact purely formal: internal lines are off-shell (do not satisfy $p^{2} = m^{2}$ ), virtual particles propagate "backward in time," and the same set of rules yields the famous agreement of QED with experiment to twelve significant figures.

The Core Insight

The perturbative expansion of any $n$ -point Green's function is a sum over contractions (Wick's theorem). A diagram is a graphical encoding of one contraction pattern. Because the structure of these patterns is dictated entirely by the free propagator and the interaction vertices in the Lagrangian, one can read off — directly from $L$ — a finite set of rules that generate every term without ever re-deriving Wick's theorem. The Feynman rules thus convert dynamics (the Lagrangian) into combinatorics plus integration.

Formal Statement

Generating Functional and Perturbative Expansion

For a scalar theory with action $S [ϕ] = S_{0} [ϕ] + S_{int} [ϕ]$ , the generating functional of correlation functions is

Z [J] = \int D ϕ exp (i S_{0} [ϕ] + i S_{int} [ϕ] + i \int d^{4} x J (x) ϕ (x)),

⟳Insert the source term

J ϕ

into the path integral; functional derivatives

δ / δ J

bring down fields, generating correlation functions.

with the interacting $n$ -point function obtained by

⟨ 0∣ T ϕ (x_{1}) \dots ϕ (x_{n}) ∣0 ⟩ = \frac{1}{Z [ 0 ]} \frac{δ ^{n} Z [ J ]}{i ^{n} δ J ( x _{1} ) \dots δ J ( x _{n} )}_{J = 0} .

⟳Each functional derivative pulls a factor

ϕ (x_{k})

from the exponential source term; setting

J = 0

removes the rest.

Here $ϕ$ is the field, $S_{0}$ the free (quadratic) action, $S_{int}$ the interaction (e.g. $- \frac{λ}{4 !} ϕ^{4}$ ), $J$ an external source, $T$ time-ordering, and $∣0 ⟩$ the free vacuum.

The Position-Space Feynman Rules (scalar $\phi^4$)

Expanding $exp (i S_{int})$ in powers of $λ$ and applying Wick's theorem yields a sum of products of free propagators. The rules are:

Propagator (internal or external line between

x, y

D_{F} (x - y) = \int \frac{d ^{4} k}{( 2 π ) ^{4}} \frac{i}{k ^{2} - m ^{2} + i ϵ} e^{- ik \cdot (x - y)} .

⟳Invert the quadratic operator

(□ + m^{2})

; the

i ϵ

enforces Feynman (time-ordered) boundary conditions via the pole prescription.

Vertex: each interaction

- \frac{λ}{4 !} ϕ^{4}

contributes a factor

- iλ

and an integration

\int d^{4} z

over the vertex location, joining four lines. Symmetry factor: divide by the order of the diagram's automorphism group

∣ Aut (Γ) ∣

Momentum-Space Feynman Rules

For amputated, momentum-space amplitudes the rules become purely algebraic. For a scalar with $ϕ^{n}$ interaction:

Internal line of momentum $k$ : $\frac{i}{k ^{2} - m ^{2} + i ϵ}$ .
Vertex: $- iλ (2 π)^{4} δ^{4} (\sum k_{in} - \sum k_{out})$ (momentum conservation).
Loop integration: $\int \frac{d ^{4} k}{( 2 π ) ^{4}}$ for each independent loop.
External legs (amputated): factor $1$ for scalars; appropriate wavefunctions $u, \overset{u}{ˉ}, v, \overset{v}{ˉ}, ϵ^{μ}$ for fermions/photons.
Symmetry factor $1/∣ Aut (Γ) ∣$ and a sign $(- 1)$ for each closed fermion loop.

The connected, amputated $n$ -point amplitude $i M$ is the sum over all topologically distinct diagrams.

i M (2 \to 2)_{tree, ϕ^{4}} = - iλ .

⟳At tree level the single 4-point vertex contributes

- iλ

; external lines are amputated (no propagators), and there is no loop integration.

Why It Holds / Derivation

From the Dyson Series to Diagrams

Step 1 — Interaction picture. Write the S-matrix as the Dyson series

S = T exp (- i \int_{- \infty}^{\infty} d t H_{int} (t)) = n = 0 \sum \infty \frac{( - i ) ^{n}}{n !} \int d^{4} x_{1} \dots d^{4} x_{n} T [H_{int} (x_{1}) \dots H_{int} (x_{n})] .

⟳Solve the interaction-picture Schrödinger equation

i \partial_{t} U = H_{int} U

iteratively; time-ordering arises from nesting the integration limits.

Step 2 — Wick's theorem. Each time-ordered product of free fields equals the sum of all normal-ordered terms with every possible set of pairwise contractions:

T [ϕ (x_{1}) \dots ϕ (x_{n})] = contractions \sum : (uncontracted fields) : pairs \prod D_{F} (x_{i} - x_{j}) .

⟳Induct on the number of fields; each new field either remains uncontracted (normal-ordered) or contracts with one of the existing fields, giving a propagator.

Step 3 — Vacuum matrix element. Sandwiching between

⟨ 0∣ \dots ∣0 ⟩

kills all normal-ordered terms with surviving fields; only fully contracted terms (where external operators pair with internal vertex fields) survive. Step 4 — Combinatorics. The

1/ n!

from the exponential is cancelled by the

n!

ways of permuting identical vertices; the

1/4!

in the

ϕ^{4}

coupling is (mostly) cancelled by the

4!

ways of attaching four lines to a vertex. Residual uncancelled factors are precisely the symmetry factor

1/∣ Aut (Γ) ∣

. Step 5 — Fourier transform. Replacing each

D_{F}

by its momentum representation (Eq. 3) and performing the

\int d^{4} x

at each vertex generates the momentum-conserving delta function (rule 2). Stripping one overall

(2 π)^{4} δ^{4} (\sum p)

defines

M

This establishes that the diagrammatic rules reproduce, term by term, the Dyson series. $■$

Richard P. Feynman | 1918–1988

Feynman introduced his diagrams at the 1948 Pocono Conference, initially to a skeptical audience that found his intuitive, "particle-trajectory" picture unjustified compared to Schwinger's and Tomonaga's operator formalisms. His genius was to recognize that the antiparticle could be treated as a particle moving backward in time (the Stückelberg–Feynman interpretation), unifying particle and antiparticle propagation in a single propagator with the $i ϵ$ prescription. He developed the rules as a physicist's calculational tool; it fell to Freeman Dyson (1949) to prove their equivalence to the Schwinger–Tomonaga theory and embed them in the systematic Dyson series. Feynman shared the 1965 Nobel Prize with Schwinger and Tomonaga.

Building Intuition

Think of a correlation function as asking: "Given a disturbance at points $x_{1}, \dots, x_{n}$ , what is the amplitude for the field to respond?" The field propagates between points along every possible path, scattering at interaction vertices. A diagram is one such history; the full amplitude sums over all histories. Internal lines are virtual — they need not satisfy $E^{2} = p^{2} + m^{2}$ because they exist only fleetingly, protected by the energy-time uncertainty relation. The $i$ 's and signs are not arbitrary: the $i$ in each propagator and vertex tracks the $i$ in $exp (i S)$ , while the $i ϵ$ encodes causality by selecting the time-ordered (Feynman) boundary conditions. Loop integrations $\int d^{4} k$ represent summing over all momenta the unobserved virtual particle can carry — the origin of both quantum corrections and ultraviolet divergences.

Tree-level vs. one-loop diagrams for $2\to2$ scattering in $\phi^4$ theory

- iλ

Worked Examples

Tree-level Compton-like scattering rate in scalar QED

Given: Scalar Yukawa theory

L_{int} = - g ϕ \overset{ˉ}{ψ} ψ

, computing the tree amplitude for fermion–fermion scattering

ψ (p_{1}) ψ (p_{2}) \to ψ (p_{3}) ψ (p_{4})

via single scalar exchange. Steps:

Two diagrams (t-channel and u-channel). For the t-channel:

- Two vertices: each

- i g

. - Internal scalar propagator with momentum

q = p_{1} - p_{3}

\frac{i}{q ^{2} - m _{ϕ}^{2}}

. - External fermion spinors:

\overset{u}{ˉ} (p_{3})

u (p_{1})

\overset{u}{ˉ} (p_{4})

u (p_{2})

Assemble:

i M_{t} = (- i g)^{2} \frac{i}{( p _{1} - p _{3} ) ^{2} - m _{ϕ}^{2}} [\overset{u}{ˉ} (p_{3}) u (p_{1})] [\overset{u}{ˉ} (p_{4}) u (p_{2})] .

⟳Read the diagram from external lines inward; each vertex contributes

- i g

and a propagator joins the two vertices.

Numerically, take $g = 1$ , $m_{ϕ} = 0$ , and a momentum transfer $t \equiv (p_{1} - p_{3})^{2} = - (0.5 GeV)^{2} = - 0.25 GeV^{2}$ . The propagator factor is $i / (- 0.25 GeV^{2}) = - 4 i GeV^{- 2}$ .

Boxed answer:

M_{t} = - 4 [\overset{u}{ˉ}_{3} u_{1}] [\overset{u}{ˉ}_{4} u_{2}] GeV^{- 2}

(per unit spinor bilinear), with the full amplitude being

M_{t} - M_{u}

for identical fermions.

One-loop scalar self-energy and its symmetry factor

Given:

ϕ^{4}

theory,

L_{int} = - \frac{λ}{4 !} ϕ^{4}

. Compute the one-loop correction to the propagator (the "tadpole" self-energy). Steps:

One vertex $- iλ$ , with two of the four legs joined into a loop.
Symmetry factor: the loop can be contracted in $\frac{1}{2}$ (the two loop legs are interchangeable), so $S = 2$ , giving factor $1/2$ .
The amplitude:

- i Π = \frac{- iλ}{2} \int \frac{d ^{4} k}{( 2 π ) ^{4}} \frac{i}{k ^{2} - m ^{2} + i ϵ} .

⟳One vertex closes two legs into a loop; the factor

1/2

is the automorphism of swapping the loop's two attachment points.

This integral is quadratically divergent. In dimensional regularization with $d = 4 - 2 ϵ$ ,

Π = \frac{λ m ^{2}}{32 π ^{2}} (\frac{1}{ϵ} - γ_{E} + ln \frac{4 π μ ^{2}}{m ^{2}} + 1) + O (ϵ) .

⟳Use the standard tadpole integral

\int d^{d} k / (2 π)^{d} 1/ (k^{2} - m^{2}) = - i m^{d - 2} Γ (1 - d /2) / (4 π)^{d /2}

, then expand near

d = 4

Numerically, take $λ = 0.1$ , $m = 100$ GeV, $μ = 100$ GeV: the finite part is $Π_{fin} = \frac{0.1 \times 1 0 ^{4}}{32 π ^{2}} (0 - 0.5772 + ln (4 π) + 1) GeV^{2} \approx 3.167 \times (1.954) GeV^{2}$ .

Boxed answer:

Π_{fin} \approx 6.19 GeV^{2}

(the

1/ ϵ

pole is absorbed by mass renormalization).

Common Misconceptions

Diagrams are not pictures of real trajectories. Internal lines are off-shell mathematical objects; nothing literally travels along them. The "virtual particle" language is a heuristic. Likewise, the apparent "speed" or "direction in time" of internal lines has no observable meaning — only the total amplitude, summed over all diagrams at a given order, is physical. The symmetry factor is easy to get wrong. It is

1/∣ Aut (Γ) ∣

, the order of the graph's automorphism group treating identical lines and vertices as interchangeable while keeping external legs fixed and labeled. For tree diagrams with distinct external legs the factor is usually 1, but loops, self-contractions (tadpoles), and multiple identical internal lines generate factors of

2, 4, 6, \dots

Always recount by explicit Wick contraction when in doubt. Naively, one might think every line in a diagram satisfies $p^{2} = m^{2}$ . Consider the s-channel one-loop bubble: if the loop momentum

k

were forced on-shell, the integral would collapse to a finite phase-space integral. But the loop momentum ranges over all of

R^{4}

, including

k^{2} \neq = m^{2}

. This off-shell region is precisely what makes loop integrals divergent and necessitates renormalization — a phenomenon invisible if one (incorrectly) restricts to on-shell internal lines.

Connections

The Feynman rules are inseparable from the broader edifice of perturbative QFT. They presuppose Wick's Theorem for converting time-ordered products into contractions, and the Path Integral Formulation or the Dyson Series for their derivation. Loop diagrams generated by the rules require Regularization and Renormalization to render divergent integrals finite and physical. The rules for gauge theories demand Faddeev-Popov Ghosts and gauge fixing; their consistency is guaranteed by BRST Symmetry and the Ward-Takahashi Identities. At higher orders, the structure of loop integrals connects to Dimensional Regularization and modern Amplitude Methods such as unitarity cuts and the Optical Theorem.

Real-World Applications

Anomalous magnetic moment of the electron $(g - 2)$ : QED Feynman diagrams through five loops predict $a_{e}$ to twelve significant figures, the most precise theory-experiment agreement in physics.
LHC cross sections: NLO and NNLO QCD computations for Higgs production, top-pair production, and jet rates rely entirely on automated Feynman-rule-based amplitude generators (MadGraph, FeynArts, GoSam).
Electroweak precision tests: Predictions for the $W$ and $Z$ boson masses, the $ρ$ parameter, and $Z$ -pole observables at LEP used multi-loop Feynman diagrams to constrain the top and Higgs masses before discovery.
Neutrino oscillation amplitudes: Charged- and neutral-current interaction vertices give the matrix elements for neutrino scattering off nuclei in detectors.
Lamb shift: The splitting of the $2 S_{1/2}$ and $2 P_{1/2}$ hydrogen levels is computed from electron self-energy and vacuum-polarization diagrams.
Casimir and vacuum-energy effects in QED arise from vacuum-polarization (photon self-energy) diagrams.
Lattice-matched perturbation theory: Continuum Feynman rules provide the matching coefficients (Wilson coefficients) connecting lattice QCD results to physical observables.

Historical Development

1929–1932: Heisenberg and Pauli formulate QED; divergences plague all calculations.
1937: Stückelberg interprets antiparticles as particles moving backward in time, prefiguring the propagator.
1947: The Lamb shift and electron $g - 2$ anomaly are measured; theory urgently needs a finite calculational scheme.
1948 (Pocono Conference): Feynman presents his diagrams; Schwinger and Tomonaga present their operator-based renormalization. Feynman's pictorial method is initially met with skepticism.
1949: Dyson publishes "The Radiation Theories of Tomonaga, Schwinger, and Feynman," proving their equivalence and deriving Feynman's rules systematically from the Dyson series.
1949: Feynman publishes "Space-Time Approach to Quantum Electrodynamics," codifying the rules.
1965: Feynman, Schwinger, and Tomonaga share the Nobel Prize in Physics.
1967–1972: 't Hooft and Veltman extend rules to non-Abelian gauge theories, introducing Faddeev–Popov ghosts and dimensional regularization, proving renormalizability.
1990s–present: Automation (FeynArts, FORM, MadGraph) and on-shell/unitarity methods transform multi-loop computation, while the rules remain the conceptual foundation.

Subtleties & Limits

The Feynman rules generate a divergent asymptotic series, not a convergent one: the number of diagrams grows factorially with order, and the perturbative expansion has zero radius of convergence (Dyson's argument). Summing the series requires Borel resummation, which encounters renormalons and instanton singularities encoding non-perturbative physics invisible to any finite set of diagrams — confinement and chiral symmetry breaking in QCD being prime examples. Furthermore, the standard rules assume a well-defined asymptotic Fock space, which fails for theories with massless particles (infrared divergences requiring the KLN theorem and soft/collinear resummation) and for genuinely strongly coupled or background-dependent settings (gravity, de Sitter space). Modern approaches — the Amplitude Methods revolution, on-shell recursion, and the amplituhedron — suggest the Feynman-diagram expansion may be a highly redundant, non-optimal organization of amplitudes whose true structure is more economically captured by other variables.