Feynman Diagrams

Feynman diagrams are graphical representations of terms in the perturbative expansion of scattering amplitudes in quantum field theory. Each diagram encodes a specific integral over internal momenta via the Feynman rules, with vertices, propagators, and external legs. They apply wherever a weakly-coupled QFT or many-body theory admits a perturbative series — QED, QCD (perturbative regime), electroweak theory, and condensed matter. Introduced by Richard Feynman (1948–1949), with parallel formalisms by Schwinger and Tomonaga, and unified by Dyson.

A Feynman diagram is a topological graph whose edges (lines) and nodes (vertices) translate, via a fixed dictionary called the Feynman rules, into a definite contribution to a quantum amplitude $M$ at a given order in the coupling constant. Vertices represent local interactions dictated by the interaction Lagrangian; internal lines represent virtual particle propagators; external lines represent incoming/outgoing physical states. They matter because they reduce the otherwise intractable bookkeeping of Wick's theorem and the Dyson series to an organized, combinatorial, and physically transparent calculation. The surprising fact: lines in a Feynman diagram do not represent particles traveling along definite trajectories — they are mnemonic bookkeeping for momentum-space integrals, and "virtual particles" need not lie on their mass shell, $p^{2} \neq = m^{2}$ .

The Core Insight

The time-ordered exponential of the interaction Hamiltonian — the Dyson series — when evaluated between asymptotic in/out states using Wick's theorem, decomposes into a sum of contractions. Each distinct contraction pattern is a graph. Feynman's genius was to read the graph backwards: instead of computing the algebra and drawing the picture afterward, one draws all topologically distinct diagrams of a given order and reads off the amplitude directly from a small set of rules. Combinatorics becomes geometry.

Formal Statement

Perturbative S-matrix and the Dyson Series

The scattering operator (S-matrix) is the time-ordered exponential of the interaction Hamiltonian density $H_{I}$ :

S = T exp (- i \int d^{4} x H_{I} (x)) = n = 0 \sum \infty \frac{( - i ) ^{n}}{n !} \int d^{4} x_{1} \dots d^{4} x_{n} T {H_{I} (x_{1}) \dots H_{I} (x_{n})}

⟳Solve the interaction-picture equation

i \partial_{t} U = H_{I} (t) U

iteratively (Picard iteration); the time-ordering symbol

T

absorbs the

1/ n!

from symmetrizing the nested integration domains.

Symbols: $T$ = time-ordering operator (later times to the left); $H_{I}$ = interaction Hamiltonian density; $x_{i}$ = spacetime four-vectors; $n$ = perturbative order.

Feynman Rules in Momentum Space (φ⁴ theory)

For the scalar theory $L = \frac{1}{2} (\partial ϕ)^{2} - \frac{1}{2} m^{2} ϕ^{2} - \frac{λ}{4 !} ϕ^{4}$ , the amputated amplitude $i M$ is assembled from:

Propagator: \frac{i}{p ^{2} - m ^{2} + i ϵ}, Vertex: - iλ (2 π)^{4} δ^{4} (\sum p_{in} - \sum p_{out})

⟳The propagator is the Fourier transform of the Feynman two-point function

⟨ 0∣ T ϕ (x) ϕ (y) ∣0 ⟩

; the vertex factor

- iλ

comes from one power of the interaction term, with

4! = 24

canceling the

1/4!

for distinct leg assignments.

Each independent loop carries an integral $\int \frac{d ^{4} k}{( 2 π ) ^{4}}$ ; momentum is conserved at each vertex; divide by the symmetry factor of the diagram.

Symbols: $p$ = four-momentum, $m$ = mass, $λ$ = coupling, $k$ = loop momentum, $ϵ \to 0^{+}$ = Feynman pole prescription.

Wick's Theorem

Any time-ordered product of free fields equals the sum over all possible full and partial contractions, normal-ordered:

T {ϕ (x_{1}) \dots ϕ (x_{n})} = contractions \sum : ϕ (x_{i_{1}}) \dots \contraction ϕ (x_{a}) ϕ ϕ (x_{a}) ϕ (x_{b}) \dots :

where each contraction $\contraction ϕ ϕ ϕ (x_{a}) ϕ (x_{b}) = D_{F} (x_{a} - x_{b})$ is the Feynman propagator.

⟳Decompose each field into positive/negative frequency parts,

ϕ = ϕ^{+} + ϕ^{-}

, and induct on the number of fields, commuting annihilation operators rightward; each commutator generates a propagator.

Linked-Cluster / Connected Diagrams

The generating functional satisfies $Z [J] = e^{iW [J]}$ , where $W [J]$ generates only connected diagrams. Consequently disconnected vacuum bubbles exponentiate and cancel in normalized correlation functions:

⟨ Ω∣ T {ϕ (x_{1}) \dots ϕ (x_{n})} ∣Ω ⟩ = \frac{⟨ 0∣ T { ϕ \dots ϕ e ^{- i \int H_{I}} } ∣0 ⟩}{⟨ 0∣ T { e ^{- i \int H_{I}} } ∣0 ⟩}

⟳Factor every diagram into connected pieces; the sum over all bubble configurations factorizes into

exp (\sum connected bubbles)

, which is precisely the denominator.

Why It Holds / Derivation

From Dyson Series to Diagrams

Step 1. Write a generic S-matrix element

⟨ f ∣ S ∣ i ⟩

between asymptotic Fock states, inserting the expansion [1]. Step 2. At each order

n

, the matrix element is a vacuum expectation value of a time-ordered product of fields (from

H_{I}

) and the creation/annihilation operators defining

∣ i ⟩, ∣ f ⟩

. Step 3. Apply Wick's theorem [3]. Fully-contracted terms survive between the vacuum; each contraction is a propagator

D_{F}

, and contractions with external operators give external legs. Step 4. Catalogue the contraction patterns topologically. Each distinct pattern ↔ one Feynman diagram. Vertices ↔ insertions of

H_{I}

; internal lines ↔ internal contractions; external lines ↔ contractions with

∣ i ⟩, ∣ f ⟩

. Step 5. Count combinatorics. The

1/ n!

from [1] is cancelled by the

n!

ways of assigning vertices to a labeled diagram, except for residual symmetry: when a diagram is invariant under a permutation of its lines/vertices, divide by the symmetry factor

S

. Step 6. Fourier transform to momentum space. Each vertex integral

\int d^{4} x_{i}

produces a momentum-conserving delta. One overall delta enforces total conservation; the rest fix internal momenta, leaving

L = I - V + 1

unconstrained loop integrals. This yields rule [2].

■

Richard P. Feynman | 1918–1988

Feynman introduced his diagrams at the 1948 Pocono Conference, initially to skepticism from Bohr and others who thought he was reviving naive spacetime trajectories. His 1949 papers "Space-Time Approach to Quantum Electrodynamics" and "The Theory of Positrons" established the rules and the interpretation of antiparticles as negative-energy solutions propagating backward in time. He shared the 1965 Nobel Prize with Julian Schwinger and Sin-Itiro Tomonaga. Freeman Dyson proved the equivalence of Feynman's intuitive diagrams with the operator formalisms of Schwinger and Tomonaga, giving the method its rigorous footing.

Building Intuition

Think of a scattering amplitude as a sum over stories of how an initial state could become a final state. Each story is a diagram. The propagator is the amplitude for a disturbance to travel between two interaction events; its pole at $p^{2} = m^{2}$ says real particles propagate freely, but off-shell ("virtual") excitations contribute too, suppressed by how far $p^{2}$ is from $m^{2}$ . Vertices are where the coupling acts — each one costs a factor of the coupling constant, so higher-order diagrams (more vertices) are smaller when the coupling is weak. The total amplitude is the coherent sum of all stories: quantum mechanics adds amplitudes, not probabilities, so diagrams interfere.

Tree-level and one-loop diagrams for 2→2 scattering in φ⁴ theory. The s-channel tree diagram (gold) dominates; the one-loop bubble is the leading correction.

<svg viewBox="0 0 600 320" xmlns="http://www.w3.org/2000/svg">
<rect width="600" height="320" fill="#0a0e17"/>
<text x="150" y="30" fill="#e2e8f0" font-size="16" text-anchor="middle">Tree level (s-channel)</text>
<text x="450" y="30" fill="#e2e8f0" font-size="16" text-anchor="middle">One loop (bubble)</text>

<!-- Tree diagram -->
<line x1="60" y1="70" x2="130" y2="130" stroke="#94a3b8" stroke-width="2"/>
<line x1="60" y1="190" x2="130" y2="130" stroke="#94a3b8" stroke-width="2"/>
<line x1="130" y1="130" x2="200" y2="130" stroke="#d4a853" stroke-width="3"/>
<line x1="200" y1="130" x2="270" y2="70" stroke="#94a3b8" stroke-width="2"/>
<line x1="200" y1="130" x2="270" y2="190" stroke="#94a3b8" stroke-width="2"/>
<circle cx="130" cy="130" r="4" fill="#d4a853"/>
<circle cx="200" cy="130" r="4" fill="#d4a853"/>
<text x="165" y="120" fill="#d4a853" font-size="13" text-anchor="middle">k</text>
<text x="55" y="65" fill="#e2e8f0" font-size="13" text-anchor="end">p₁</text>
<text x="55" y="200" fill="#e2e8f0" font-size="13" text-anchor="end">p₂</text>
<text x="278" y="65" fill="#e2e8f0" font-size="13">p₃</text>
<text x="278" y="200" fill="#e2e8f0" font-size="13">p₄</text>

<!-- One loop diagram -->
<line x1="360" y1="70" x2="420" y2="130" stroke="#94a3b8" stroke-width="2"/>
<line x1="360" y1="190" x2="420" y2="130" stroke="#94a3b8" stroke-width="2"/>
<line x1="480" y1="130" x2="540" y2="70" stroke="#94a3b8" stroke-width="2"/>
<line x1="480" y1="130" x2="540" y2="190" stroke="#94a3b8" stroke-width="2"/>
<ellipse cx="450" cy="130" rx="30" ry="22" fill="none" stroke="#d4a853" stroke-width="3"/>
<circle cx="420" cy="130" r="4" fill="#d4a853"/>
<circle cx="480" cy="130" r="4" fill="#d4a853"/>
<text x="450" y="100" fill="#d4a853" font-size="12" text-anchor="middle">k</text>
<text x="450" y="172" fill="#d4a853" font-size="12" text-anchor="middle">k−p</text>
<text x="355" y="65" fill="#e2e8f0" font-size="13" text-anchor="end">p₁</text>
<text x="355" y="200" fill="#e2e8f0" font-size="13" text-anchor="end">p₂</text>
<text x="548" y="65" fill="#e2e8f0" font-size="13">p₃</text>
<text x="548" y="200" fill="#e2e8f0" font-size="13">p₄</text>

<text x="300" y="290" fill="#94a3b8" font-size="13" text-anchor="middle">amplitude = sum of all topologically distinct diagrams</text>
</svg>

Worked Examples

Tree-level 2→2 scattering in φ⁴ theory

Given: Compute

i M

for

ϕϕ \to ϕϕ

at lowest order (single vertex), coupling

λ

. Step 1. The leading diagram has one vertex connecting all four external legs. There are no internal propagators and no loops. Step 2. Apply rule [2]: the single vertex contributes

- iλ

. The symmetry factor is

S = 1

(all four legs are distinguishable external states). Step 3. Therefore:

i M = - iλ ⟹ M = - λ

⟳With only one vertex and no internal lines, the amplitude is just the vertex factor; no loop integrals occur.

Step 4. The differential cross section in the CM frame for identical bosons:

\frac{d σ}{d Ω} = \frac{∣ M ∣ ^{2}}{64 π ^{2} s} = \frac{λ ^{2}}{64 π ^{2} s}

⟳Use the standard

2 \to 2

CM cross-section formula

d σ / d Ω = ∣ M ∣^{2} / (64 π^{2} s)

with

∣ p_{f} ∣ = ∣ p_{i} ∣

for equal masses.

Boxed answer:

M = - λ, \frac{d σ}{d Ω} = \frac{λ ^{2}}{64 π ^{2} s} [GeV^{- 2}]

where

s

is the Mandelstam invariant in GeV².

One-loop bubble and its divergence

Given: Evaluate the s-channel one-loop bubble correction to

ϕϕ \to ϕϕ

. External momentum flowing through the loop is

p = p_{1} + p_{2}

, with

p^{2} = s

. Step 1. Apply the rules: two vertices give

(- iλ)^{2}

, the symmetry factor of the bubble is

S = 2

, and there is one loop integral:

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

⟳Two propagators share the loop momentum

k

and

k - p

; the factor

1/2

is the symmetry factor for exchanging the two internal lines.

Step 2. Combine denominators with a Feynman parameter

x

and shift

k \to k + x p

M_{loop} = \frac{λ ^{2}}{2} \int_{0}^{1} d x \int \frac{d ^{4} k}{( 2 π ) ^{4}} \frac{1}{[ k ^{2} - Δ + i ϵ ] ^{2}}, Δ = m^{2} - x (1 - x) s

⟳Use

\frac{1}{A B} = \int_{0}^{1} \frac{d x}{[ x A + ( 1 - x ) B ] ^{2}}

, then complete the square in

k

Step 3. In

d = 4 - 2 ε

dimensions (dimensional regularization), the integral gives:

M_{loop} = \frac{λ ^{2}}{32 π ^{2}} \int_{0}^{1} d x [\frac{1}{ε} - γ_{E} + ln 4 π - ln \frac{Δ}{μ ^{2}}]

⟳Wick-rotate to Euclidean space (

k^{0} \to i k_{E}^{0}

) and use the master formula

\int \frac{d ^{d} k}{( 2 π ) ^{d}} \frac{1}{( k ^{2} + Δ ) ^{2}} = \frac{Γ ( 2 - d /2 )}{( 4 π ) ^{d /2}} Δ^{d /2 - 2}

Boxed answer:

M_{loop} = \frac{λ ^{2}}{32 π ^{2}} [\frac{1}{ε} + finite]

— the

1/ ε

pole signals a UV divergence absorbed by coupling renormalization.

Common Misconceptions

Feynman diagrams are not pictures of particle trajectories. The internal lines are propagators integrated over all internal momenta; "virtual particles" are off-shell (

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

) and are not directly observable. A diagram does not depict what happened — only the full coherent sum of all diagrams has physical meaning, and individual diagrams are gauge-dependent in non-abelian theories. The perturbation series is asymptotic, not convergent. For QED the coefficients eventually grow factorially; summing all orders does not give a finite answer. The series is useful because early terms shrink while

α \approx 1/137

is small, but it must be truncated. At strong coupling (e.g., low-energy QCD) the diagrammatic expansion fails entirely.

A naïve student might count a single labeled vacuum-polarization bubble and forget the symmetry factor $S = 2$ , overcounting the amplitude by a factor of two. Likewise, drawing the "tadpole" diagram and ignoring that it can be cancelled by a counterterm (or vanishes by normal-ordering) leads to spurious infinities. Symmetry factors and the connected/amputated prescriptions are not optional bookkeeping — omitting them yields the wrong cross section.

Connections

Feynman diagrams sit at the crossroads of several frameworks. They are the perturbative face of the Path Integral Formulation, where the generating functional $Z [J]$ produces them by functional differentiation. They implement Wick's Theorem graphically and rely on Renormalization to render loop integrals finite. In condensed matter they reappear as the diagrammatics of Many-Body Perturbation Theory (Matsubara/Green's-function techniques). Their UV divergences are organized by the Renormalization Group, and at high orders their factorial growth connects to Resurgence and Trans-series and non-perturbative instanton effects. Modern amplitude methods (BCFW Recursion, the amplituhedron) bypass diagrams entirely while computing the same quantities.

Real-World Applications

Anomalous magnetic moment of the electron ( $g - 2$ ): QED Feynman diagrams to five loops predict $a_{e}$ to twelve significant figures, the most precisely verified prediction in physics.
LHC collider phenomenology: NLO and NNLO Feynman-diagram computations (e.g., via automated tools like MadGraph, FORM) underpin every Standard Model background estimate at ATLAS and CMS.
Quantum electrodynamics of atoms: Lamb shift and hyperfine structure calculations rely on radiative-correction diagrams.
Deep inelastic scattering & QCD: Parton-level cross sections computed diagrammatically feed factorization theorems for proton structure.
Electroweak precision tests: W and Z boson self-energy diagrams (oblique parameters $S, T, U$ ) constrain new physics.
Condensed matter / superconductivity: The diagrammatic summation of particle-particle ladders (Cooper channel) underlies BCS theory; the GW approximation in materials science sums self-energy diagrams.
Casimir force and vacuum energy: Vacuum-bubble diagrams quantify zero-point energy shifts between conducting plates.

Historical Development

1927–1932 — Dirac formulates QED and predicts antiparticles; Heisenberg & Pauli develop field quantization, but calculations are plagued by infinities.
1947 — Lamb–Retherford measure the Lamb shift; Kusch & Foley measure the electron anomalous moment, demanding a finite computational scheme.
1947–1948 — Bethe computes a non-relativistic Lamb shift; Schwinger and Tomonaga develop covariant operator renormalization independently.
1948 (Pocono Conference) — Feynman presents his diagrammatic spacetime approach; the audience is largely unconvinced.
1949 — Feynman publishes "Space-Time Approach to QED" and "The Theory of Positrons," establishing the rules.
1949 — Dyson's paper "The Radiation Theories of Tomonaga, Schwinger, and Feynman" proves their equivalence and systematizes the rules to all orders.
1950s — Dyson and Salam establish the power-counting theory of renormalizability; the method spreads to all of particle physics.
1965 — Feynman, Schwinger, Tomonaga share the Nobel Prize.
1972 — 't Hooft and Veltman introduce dimensional regularization, taming non-abelian loop diagrams and enabling the renormalization of the electroweak theory.

Subtleties & Limits

The perturbative series defined by summing all Feynman diagrams is generically divergent (asymptotic). Its large-order behavior is governed by factorially-growing contributions whose Borel transform has singularities — renormalons and instantons — encoding genuinely non-perturbative physics inaccessible to any finite set of diagrams. Whether and how the full non-perturbative amplitude can be reconstructed from the diagrammatic series (the program of resurgence) remains an active research frontier. Separately, for strongly-coupled theories (low-energy QCD, certain condensed-matter systems) no diagrammatic expansion converges, motivating lattice methods and the gauge/gravity (holographic) correspondence. Even at the level of bookkeeping, the number of diagrams grows factorially with loop order, making high-precision computations a major computational challenge addressed by recursion relations, unitarity cuts, and integration-by-parts identities.