← Physics
Graduatequantum

Feynman Diagrams

A notation so good it looks like a cartoon: dots for interactions, lines for particles, and every scribble a shorthand for a genuine term in the perturbative expansion — read correctly, the picture and the algebra say exactly the same thing.

Before this, you should know:

The previous topic left a real mess sitting on the table. The Dyson series,

S^=1+i∫d4x  Lint(x)+i22!∫d4x1 d4x2  T[Lint(x1)Lint(x2)]+⋯ ,\hat S = 1 + i\int d^4x\;\mathcal{L}_{\text{int}}(x) + \frac{i^2}{2!}\int d^4x_1\,d^4x_2\; T\big[\mathcal{L}_{\text{int}}(x_1)\mathcal{L}_{\text{int}}(x_2)\big] + \cdots,

is correct, but every single term is a nightmare of bookkeeping. Each factor of Lint=−λ4!ϕ4\mathcal{L}_{\text{int}}=-\frac{\lambda}{4!}\phi^4 expands, via the mode expansion of ϕ^\hat\phi, into a long sum of creation and annihilation operators for every momentum mode, and computing even the second-order term means tracking which of those operators annihilate which incoming particles, which create which outgoing particles, and which simply pair up with each other and cancel out of the final answer entirely. By third order the algebra is, for a human being working by hand, close to hopeless — not because the physics is unclear, but because there are so many ways for the operators to pair up that keeping an inventory by symbols alone becomes the bottleneck.

Richard Feynman's fix was not a new piece of physics. It was a notation — and one of the most consequential notations ever invented, because it turns an error-prone algebraic bookkeeping problem into something you can see. Every term in the Dyson series, it turns out, corresponds to exactly one diagram, built from a small, fixed vocabulary of pieces: dots and lines. Learn the translation once, and you can write down (or read off) any term in the perturbative expansion without touching the underlying operator algebra directly.

The vocabulary: three kinds of mark on the page

A Feynman diagram for the ϕ4\phi^4 theory of the previous topic is built from exactly two kinds of ingredient.

Vertices. A dot, where four lines meet, represents one factor of the interaction −λ4!ϕ4-\frac{\lambda}{4!}\phi^4 evaluated at one spacetime point, integrated over d4xd^4x (i.e., over every place and time it could have happened). Each vertex costs one power of the coupling constant λ\lambda — so counting vertices in a diagram is exactly counting what order of perturbation theory it belongs to. A diagram with one vertex is a first-order (λ1\lambda^1) contribution; two vertices, second order (λ2\lambda^2); and so on, in exact correspondence with the powers of λ\lambda in the series above.

External lines. A line that begins or ends at the edge of the diagram, not at another vertex, represents a real, physical particle — one that is actually present in the initial or final state being computed, with a definite, on-shell momentum satisfying the dispersion relation ωk=k2+m2\omega_k=\sqrt{k^2+m^2} from quantum field theory. An external line entering the diagram is a creation or annihilation operator from the mode expansion of ϕ^\hat\phi acting directly on the initial or final Fock-space state — the literal operator algebra from the free-field topic, simply drawn as a line instead of written as a^k†∣0⟩\hat a_k^\dagger|0\rangle.

Internal lines. A line connecting two vertices to each other, touching the edge of the diagram nowhere, represents a particle that is created at one vertex and destroyed at another — a a^k†\hat a_k^\dagger from one factor of ϕ^(x1)\hat\phi(x_1) paired algebraically with an a^k\hat a_k from a factor of ϕ^(x2)\hat\phi(x_2) at a different vertex, summed over every possible intermediate momentum kk it could have carried. Such a particle is called virtual: it is never measured, never part of the initial or final state, and — this is the genuinely strange part, worth sitting with rather than rushing past — it does not have to satisfy ωk=k2+m2\omega_k=\sqrt{k^2+m^2} the way a real particle must. It exists only as an intermediate step in the algebra, a bookkeeping device for "a quantum was created here and annihilated there," and its mathematical description (the propagator) is the precise field-theoretic descendant of the same creation-and-annihilation-operator language built for a single spring in the quantum harmonic oscillator and extended to a whole field in quantum field theory.

A single Feynman diagram: two straight lines enter from the left labeled k1 and k2, meeting at a single dot in the center labeled with a factor of minus i lambda, from which two straight lines emerge to the right labeled k3 and k4. No internal lines are present, only the one vertex and four external legs.

The simplest possible interaction diagram in φ⁴ theory: one vertex, four external legs, zero internal lines. Two particles come in, meet at a single point of interaction, and two particles come out.

Worked example

Draw the simplest possible Feynman diagram for two-to-two scattering in ϕ4\phi^4 theory, and show explicitly that the interaction term in the Dyson series produces exactly this process while enforcing four-momentum conservation at the vertex. (click to reveal the solution)

Setting up: take an initial state of two particles with momenta k1,k2k_1,k_2 and ask for the amplitude to find two particles with momenta k3,k4k_3,k_4 in the final state, at first order in λ\lambda — a single insertion of the interaction:

∣i⟩=a^k1†a^k2†∣0⟩,⟨f∣=⟨0∣a^k4a^k3,S^(1)=−iλ4!∫d4x  ϕ^(x)4.|i\rangle = \hat a_{k_1}^{\dagger}\hat a_{k_2}^{\dagger}|0\rangle, \qquad \langle f| = \langle 0|\hat a_{k_4}\hat a_{k_3}, \qquad \hat S^{(1)} = -i\frac{\lambda}{4!}\int d^4x\;\hat\phi(x)^4.

We want ⟨f∣S^(1)∣i⟩\langle f|\hat S^{(1)}|i\rangle.

Expanding ϕ^(x)4\hat\phi(x)^4. Recall the mode expansion from quantum field theory, written compactly as ϕ^(x)=∑k12ωkV(a^keik⋅x+a^k†e−ik⋅x)\hat\phi(x)=\sum_k \frac{1}{\sqrt{2\omega_k V}}\big(\hat a_k e^{ik\cdot x}+\hat a_k^{\dagger}e^{-ik\cdot x}\big), with k⋅xk\cdot x shorthand for k⋅x⃗−ωktk\cdot\vec x-\omega_k t. Multiplying four copies of this together produces a sum of terms, each a product of four operators, each either an a^\hat a or an a^†\hat a^\dagger. To turn ∣i⟩|i\rangle (two particles) into ⟨f∣\langle f| (two different particles), exactly two of the four operator factors must be annihilation operators — one to remove the k1k_1 particle, one to remove the k2k_2 particle — and the other two must be creation operators, building the k3k_3 and k4k_4 particles fresh. Any term in the expansion with three annihilators and one creator, or four of one kind, gives zero when sandwiched between a two-particle initial state and a two-particle final state — there simply isn't a matching particle for the extra operator to act on.

Isolating the surviving piece. Among the surviving terms, the specific combination that connects precisely to k1,k2→k3,k4k_1,k_2\to k_3,k_4 carries the phase factors from each mode's exponential:

⟨f∣ a^k1a^k2a^k3†a^k4† ∣i⟩  ∝  eik1⋅x eik2⋅x e−ik3⋅x e−ik4⋅x,\langle f|\,\hat a_{k_1}\hat a_{k_2}\hat a_{k_3}^{\dagger}\hat a_{k_4}^{\dagger}\,|i\rangle \;\propto\; e^{ik_1\cdot x}\,e^{ik_2\cdot x}\,e^{-ik_3\cdot x}\,e^{-ik_4\cdot x},

one phase per operator, each carrying the momentum of the particle it creates or destroys, evaluated at the one shared point xx where all four lines meet — exactly the picture of a single vertex the diagram above draws. (The full calculation also carries normalization factors 1/2ωkV1/\sqrt{2\omega_kV} per leg and a combinatorial count of how many ways to assign the four operators in ϕ^4\hat\phi^4 to the four external legs — bookkeeping that reproduces the standard vertex factor of −iλ-i\lambda once every contraction is tallied. Carrying that count through in full belongs to a complete course in quantum field theory; the physics of interest here is what happens next.)

Integrating over the vertex position. The amplitude requires ∫d4x\int d^4x of the product of phases above:

∫d4x  ei(k1+k2−k3−k4)⋅x=(2π)4 δ4(k1+k2−k3−k4),\int d^4x\; e^{i(k_1+k_2-k_3-k_4)\cdot x} = (2\pi)^4\,\delta^4(k_1+k_2-k_3-k_4),

using the same Fourier representation of the delta function used to close out the canonical commutator derivation in the previous topic. This delta function is not an extra assumption bolted on afterward — it is a forced consequence of integrating a single interaction point over all of spacetime, and it says exactly one thing:

k1+k2=k3+k4.\boxed{k_1+k_2=k_3+k_4.}

Interpreting the result. Four-momentum — energy and ordinary momentum together — is conserved at the vertex, automatically, with no separate conservation law imposed by hand. This is the precise mathematical content of the diagram: two lines carrying k1,k2k_1,k_2 come in, two lines carrying k3,k4k_3,k_4 go out, the dot where they meet costs one factor of λ\lambda, and the requirement that all four lines meet at a single point of spacetime is exactly what forces the incoming total momentum to equal the outgoing total momentum. There are no internal lines in this simplest diagram — no particle is created and later destroyed inside the diagram itself — because a single vertex has exactly four legs, and all four are needed just to hold the two incoming and two outgoing particles. Internal lines, and the virtual particles they represent, first appear at second order, where two vertices are connected by one line while the rest of their eight legs remain external — a diagram genuinely worth drawing once this one is second nature, but one that needs a second vertex's worth of algebra this worked example was not built to carry.

Reading a diagram as a sentence, not a picture

Once the translation is second nature, a Feynman diagram stops being decoration and starts being read the way you'd read an equation: left to right (or bottom to top, depending on convention) as a narrative of what happened, at what order in the coupling, with momentum conserved at every single vertex along the way. Complicated processes are diagrams with more vertices and more internal lines, and — this is the entire payoff of the whole exercise — drawing every topologically distinct way to connect the given external lines with a given number of vertices is a purely combinatorial, visual task, far more tractable than expanding the corresponding operator products by hand. Each diagram translates back into a specific integral (its Feynman rules turn dots into factors of λ\lambda, external lines into normalization factors, and internal lines into propagators, all multiplied together and integrated over unconstrained internal momenta), and the sum of every diagram at a given order is that order's term in the Dyson series — nothing more mysterious than a picture-language for exactly the algebra the previous topic wrote down.

Where this leads

Everything here was built for a toy theory — a single scalar field interacting with itself — chosen because it is the simplest theory where the diagrammatic machinery can be seen clearly, without the distraction of extra particle types or spin. The real machinery of nature has more actors: an electron field, a photon field, and a specific interaction term between them fixed not by an arbitrary choice like ϕ4\phi^4 but by a deep symmetry principle. Apply exactly this same diagrammatic bookkeeping — vertices, external lines, internal propagators, momentum conservation at every vertex — to that theory, and you get quantum electrodynamics: the quantum theory of light and matter, and the most precisely tested theory in the history of physics.