Quantum Field Theory

When you try to merge quantum mechanics with special relativity, particles stop being the fundamental objects. Fields do. Every particle in the universe becomes a ripple in one of a handful of underlying quantum fields, and the rules for how ripples scatter, create each other, and annihilate are what we call QFT. It is the most accurate theoretical framework humans have ever written down.

1. Why QFT exists

Ordinary quantum mechanics handles one or two particles at a time. You write down a wavefunction $\psi(x, t)$, evolve it with the Schrödinger equation, and compute probabilities. It works beautifully for hydrogen atoms and chemical bonds. It fails the moment you push it to relativistic energies.

Two specific things break. First, Einstein's $E = mc^2$ says energy and mass are interchangeable. If you smash two electrons together hard enough, you can spit out an electron-positron pair and extra photons — the particle count is not conserved. A fixed-particle-number wavefunction has nowhere to put the new particles. Second, even for a single particle, combining quantum mechanics with relativity produces negative-energy solutions, negative probabilities, and faster-than-light propagation if you take the naive Schrödinger approach seriously.

The fix, worked out in stages from 1927 (Dirac) through 1948 (Feynman, Schwinger, Tomonaga) and onward, is to stop treating particles as fundamental. Instead, the fundamental object is a field that fills all of space. A field is just a function: at every point in spacetime, you have a number (or a vector, or a spinor). Classically, the electromagnetic field at a point tells you the electric and magnetic field values there. Quantum mechanically, that field value becomes an operator on a Hilbert space, and the states of the theory are no longer "one particle over there, another over here" but configurations of the field itself. Particles reappear as quantized excitations — discrete units of vibration in the field — like phonons in a crystal or photons in an optical cavity.

Why should you care beyond the poetry? Four reasons:

• It predicts the universe's most precise numbers. QED's prediction of the electron's magnetic moment matches experiment to about 12 decimal places. Nothing else in science comes close.
• It underlies everything at the LHC. Every Higgs-boson discovery plot, every cross-section measurement, every exclusion limit on supersymmetric particles, is a QFT calculation compared to data.
• It connects to condensed matter. The same path-integral and renormalization tools describe phase transitions, superconductors, and the fractional quantum Hall effect. See frontier physics.
• It is the right framework for thinking about anything "quantum plus fields". Quantum optics, cavity QED, analog gravity in condensed matter, the early universe — all run on QFT.

This page will walk you through the logic without drowning you in path integrals. You will leave with a working mental picture of what a field looks like when you quantize it, what a Feynman diagram actually computes, why renormalization is not cheating, and why the Standard Model of particle physics is a particular set of QFT choices.

2. Vocabulary cheat sheet

These symbols show up on every QFT page you will ever read. Skim them now; the sections below define each properly.

One warning: QFT notation is gnarly and inconsistent across textbooks. Different authors use different metric signatures, different Fourier conventions, and different sign conventions for the action. On this page we will prioritize clarity over conformity.

3. Field quantization — classical fields become operators

Think about a guitar string. Classically, it is a smooth curve $y(x, t)$ that vibrates with various frequencies. You can decompose any motion into normal modes — the fundamental and its harmonics — and the total motion is a sum over modes, each behaving like a simple harmonic oscillator.

Quantum mechanically, a simple harmonic oscillator has discrete energy levels $E_n = \hbar \omega (n + \tfrac12)$, where $n = 0, 1, 2, \dots$ is the number of quanta of excitation. A quantum string has one such ladder per mode. When you excite mode $k$ from its ground state to $n = 1$, you have added one "quantum" of vibration in that mode. If the string were the electromagnetic field, that quantum would be a photon of momentum $k$.

This is the whole idea of field quantization. Take a classical field. Decompose it into Fourier modes. Treat each mode as a quantum harmonic oscillator. The "excitations" of the modes are the particles of the theory. A field with many modes excited corresponds to many particles flying around.

The creation and annihilation operators obey commutation relations that are the whole engine of the theory:

Once you have the operators, you build the Hilbert space: the vacuum $|0\rangle$ (with $a_k|0\rangle = 0$ for every $k$), one-particle states $a^\dagger_k |0\rangle$, two-particle states $a^\dagger_{k_1} a^\dagger_{k_2} |0\rangle$, and so on. The whole thing is called Fock space. It is automatically a variable-particle-number theory, which is exactly what relativity demanded.

4. The scalar field and the Klein-Gordon equation

The simplest field you can quantize is a real-valued scalar — just one number at each spacetime point, no internal structure. It is not a realistic particle on its own (the Higgs boson is the only fundamental scalar in the Standard Model, and even it is complex-valued), but it is the cleanest teaching example, and every more elaborate field reuses the same ideas.

The classical equation of motion for a free scalar of mass $m$ is the Klein-Gordon equation:

The Lagrangian density whose Euler-Lagrange equation gives Klein-Gordon is

Interactions are terms you tack onto this free Lagrangian. A $\lambda \phi^4$ term gives you a quartic self-interaction — scalars can scatter off themselves. A $-e\, \bar\psi \gamma^\mu \psi A_\mu$ term in QED couples fermions $\psi$ to the photon field $A_\mu$. The free theory is solvable in closed form; the interacting theory is not, and you compute it as a series in small coupling constants. That is what Feynman diagrams are for.

5. Feynman diagrams — perturbation theory you can draw

When the coupling constant is small (like QED's $\alpha \approx 1/137$), you can compute scattering amplitudes as a power series in the coupling. The first term is a "tree-level" contribution, the next is a "one-loop" correction, and so on. Richard Feynman's great insight in the 1940s was that each term in this series has a pictorial interpretation: a diagram with lines and vertices.

• External lines are incoming and outgoing particles. They correspond to the particles you prepared in the detector frame and the particles you measure afterward.
• Internal lines (propagators) are virtual particles — transient disturbances in the field that are not directly observed but carry the interaction. A propagator is the mathematical answer to "given the field operator acts here, what is the amplitude for a quantum to show up there?"
• Vertices are interaction points, read directly off the interaction part of $\mathcal{L}$. A $\lambda \phi^4/4!$ term gives a four-leg vertex with factor $-i\lambda$. A QED vertex has two fermion lines and one photon line, with factor $-ie\gamma^\mu$.

To compute an amplitude, draw every topologically distinct diagram with the right external lines and the fewest vertices, translate each piece via the Feynman rules (a propagator factor for each internal line, a vertex factor for each vertex, and an integral $\int d^4 k$ for each internal loop), and sum them up. Then square to get a probability, and integrate over final-state phase space to get a cross-section.

Even so, the mental picture is useful. An electron radiates a photon and re-absorbs it — that is an electron self-energy correction, and it contributes to the electron's effective mass. An electron emits a virtual photon that splits into an electron-positron pair before recombining — that is a vacuum polarization correction, and it contributes to the running of the electromagnetic coupling. The diagrams make what would otherwise be a soul-crushing series of indices into something you can literally draw on a napkin.

6. Interactive: tree-level versus one-loop scattering

Below is a schematic of $2 \to 2$ scattering — two particles in, two particles out. The slider lets you choose how many loops to include in the amplitude. You will see the relative size of each contribution, governed roughly by a factor of $\alpha / (4\pi)$ per loop. At $\alpha \approx 1/137$, one loop is already a 0.06% correction, two loops a 4-parts-in-$10^7$ correction, and so on. Most of QED's precision comes from people computing 4 and 5-loop diagrams by hand (and now by computer algebra).

Things to notice as you drag the sliders:

• Each added loop costs roughly a factor of $\alpha/(4\pi)$ — about $6 \times 10^{-4}$ at the physical electromagnetic coupling. This is why QED is a convergent-looking series even though formally it is asymptotic.
• Crank up $\alpha$ toward $0.3$ (the rough scale of the strong coupling at intermediate energies) and the series stops converging visually. This is why you cannot do QCD perturbatively at low energies — you need lattice methods or effective theories instead.
• The labelled diagram on the left does not change structurally; only which internal-lines-and-vertices combination dominates changes. Higher-loop diagrams have more internal momentum integrals and more places for virtual particles to wander.

7. Renormalization — why the infinities are not a disaster

Compute a loop diagram and you get an integral of the form $\int d^4 k / k^2$ extending to arbitrarily large momentum $k$. That integral is infinite. QFT was almost abandoned in the 1930s because of these ultraviolet divergences. The way out, pinned down in the late 1940s, is called renormalization. It has two stages.

Stage one: regulate. Impose some upper cutoff $\Lambda$ on the momentum integral, so you are only summing contributions up to energy $\Lambda$. The integral becomes finite but $\Lambda$-dependent.

Stage two: renormalize. Re-express the theory's parameters — the mass, the charge, the field strength — in terms of the measured values instead of the "bare" values written in $\mathcal{L}$. The $\Lambda$-dependence gets absorbed into the definition of those parameters. What is left, when you compute any physical quantity, is a finite, $\Lambda$-independent prediction.

A striking consequence: the effective coupling constants you measure depend on the energy at which you are probing the theory. Electromagnetic charge grows with energy (because vacuum polarization screens the bare charge less at short distances), while the strong coupling shrinks with energy (the opposite sign, called asymptotic freedom, which was discovered by David Gross, Frank Wilczek, and Hugh Politzer in 1973 and won them the 2004 Nobel Prize). Running couplings are testable: you can measure $\alpha$ at the Z-boson mass (about 91 GeV) and find $\alpha(M_Z) \approx 1/128$, noticeably larger than the low-energy $1/137$.

For a slow introduction aimed at a wider audience, think of renormalization as "parametrizing ignorance." You do not know what happens at the Planck scale. But every measurement you make at LHC energies can be expressed using parameters defined at LHC energies, and the Planck-scale physics leaks in only through small, computable corrections suppressed by ratios like $(\text{LHC energy}/\text{Planck})^2 \sim 10^{-32}$. Renormalization is how you quarantine ignorance about what you cannot see.

8. Gauge theories and QED as the canonical example

The Standard Model's gauge theories fall out of asking: "what if I demand my Lagrangian be invariant under a local symmetry transformation?" For QED, the relevant symmetry is a change of phase of the electron field at each spacetime point independently. Writing the free Dirac Lagrangian and then requiring invariance under $\psi(x) \to e^{i\alpha(x)}\psi(x)$ forces you to introduce a new vector field $A_\mu(x)$ — the photon — that transforms as $A_\mu \to A_\mu - \partial_\mu \alpha$ to cancel the derivative of $\psi$'s phase.

Worked example: the electron-photon vertex

The interaction term in $\mathcal{L}_{\text{QED}}$ is $-e \bar\psi \gamma^\mu \psi A_\mu$. Read off the vertex: two fermion lines (one $\psi$, one $\bar\psi$) and one photon line ($A_\mu$) meet at a point, with a factor of $-i e \gamma^\mu$ attached. Every QED process is built from this single vertex repeated. Tree-level Compton scattering (an electron scatters off a photon) uses two vertices. Electron-electron scattering via photon exchange uses two vertices. The muon's anomalous magnetic moment requires vertices arranged into a triangle-with-a-loop, and is so precisely measured that it has been a potential window to new physics for two decades (the Fermilab Muon g-2 measurement in 2021 and its follow-ups have kept the tension alive; whether it is new physics or uncertain hadronic contributions is still debated as of 2025).

9. Experimental confirmation

Physics would not have spent seven decades refining QFT if it did not pay. A short tour of the most precise agreements:

• Electron anomalous magnetic moment, $g_e - 2$. The Dirac equation predicts $g_e = 2$ for a point-like spin-$1/2$ particle. QED loops add corrections. Theory, computed through five loops, and measurement (most recently the Harvard Penning trap result, 2023) agree to about one part in $10^{12}$. This is the most precise comparison of theory and experiment in all of science.
• Lamb shift. Two states of hydrogen that would be degenerate in the Dirac theory are split by about 1 GHz. Willis Lamb and Robert Retherford measured it in 1947, and Hans Bethe explained it the same summer with a prototype renormalization calculation. Lamb got the 1955 Nobel Prize; Bethe's calculation jump-started QED.
• Higgs boson discovery, 2012. ATLAS and CMS at the LHC found a new scalar at about 125 GeV. Its decay branching ratios match Standard Model predictions derived from QFT, and its properties (spin zero, CP-even) are exactly what the electroweak sector required.
• Running of $\alpha_s$. The strong coupling constant has been measured at dozens of energies by many experiments and tracks the QCD prediction from $Q \sim 2\,\text{GeV}$ up to the TeV scale.
• Jets at colliders. When quarks and gluons are produced at a collider, they "hadronize" into collimated sprays called jets. The angular distributions and energy profiles of those jets match QCD predictions to percent accuracy in regimes where perturbation theory is valid.

Not everything agrees. The muon's magnetic moment has a persistent tension at the few-sigma level; neutrino masses are not in the original Standard Model; dark matter is unaccounted for; and quantum gravity is not a renormalizable QFT in the usual sense. QFT is extremely successful but clearly not the whole story — see particle physics for the measurements and frontier physics for where things might break next.

10. A tiny numerical glimpse

Full QFT calculations require specialized software (FeynCalc, FORM, MadGraph). But a toy thermal-field computation is very much in reach. Below we compute the vacuum energy of a free scalar field on a 1D lattice — a "zero-point" sum that is finite once you put it in a box and gives intuition for the structure of QFT calculations.

import numpy as np

# ---------- Free scalar on a 1D lattice with N sites ----------
# The sum of (1/2) omega_k over all modes is the vacuum energy.
# Dispersion on a lattice of spacing a is omega_k = sqrt(m^2 + (2/a sin(k a / 2))^2).

def vacuum_energy(N=256, L=10.0, m=0.3):
    a   = L / N                         # lattice spacing
    k   = 2 * np.pi * np.fft.fftfreq(N, d=a)
    # lattice dispersion (finite-difference second derivative)
    omega = np.sqrt(m ** 2 + (2 / a * np.sin(k * a / 2)) ** 2)
    return 0.5 * omega.sum() / L      # energy density

# The answer depends on N (the cutoff). As N grows, the sum diverges
# linearly in the cutoff scale — that is the "UV divergence" you just
# heard about. Differences between two values of m are finite, though.

for N in [64, 128, 256, 512, 1024]:
    print(f"N={N:4d}   E0 = {vacuum_energy(N=N):.4f}")

# Physical observable: how does the ground-state energy shift when
# the mass changes? This difference is cutoff-independent in the limit
# of a fine lattice, and it's the kind of quantity QFT is good at.
print("Delta E (m=0.3 vs m=0.4):",
      vacuum_energy(N=1024, m=0.4) - vacuum_energy(N=1024, m=0.3))

import math

# Same computation without NumPy — slower, more explicit.

def vacuum_energy(N=256, L=10.0, m=0.3):
    a = L / N
    total = 0.0
    for j in range(N):
        k = 2 * math.pi * j / L                 # allowed lattice momenta
        omega = math.sqrt(m * m + (2 / a * math.sin(k * a / 2)) ** 2)
        total += 0.5 * omega
    return total / L

for N in [64, 128, 256]:
    print(f"N={N:3d}   E0/L = {vacuum_energy(N=N):.4f}")

Three things the output makes concrete:

• The absolute zero-point energy grows with the cutoff $N$. There is no "right" number for the vacuum energy in isolation — it depends on what you integrate up to. This is the mildest form of the cosmological constant problem.
• Differences are finite. Change $m$ and the difference $\Delta E$ converges as $N \to \infty$. Renormalization teaches you to only ever trust differences and ratios, not absolute values.
• Lattice dispersion differs from continuum dispersion. For small $k a$, the lattice formula reduces to $\sqrt{m^2 + k^2}$. For $k a$ near $\pi$, it curves over. Lattice QCD exploits exactly this structure to compute nonperturbative physics numerically.

11. Cheat sheet

See also