Path Integral Formulation QFT

The path integral formulation expresses quantum amplitudes as a sum over all possible histories of a system, weighted by $e^{i S /ℏ}$ where $S$ is the classical action. In QFT, the fundamental objects become functional integrals over field configurations, $\int D ϕ e^{i S [ϕ] /ℏ}$ , generating all correlation functions and the S-matrix. It applies to all of relativistic quantum field theory, gauge theory, statistical mechanics, and quantum gravity. Conceived by Dirac (1933), developed fully by Feynman (1948), and made rigorous in Euclidean form by later workers.

The path integral formulation recasts quantum mechanics and quantum field theory by replacing the operator-state formalism with a sum (functional integral) over all field configurations connecting a given initial and final condition, each weighted by a phase equal to the classical action divided by $ℏ$ . It matters because it makes Lorentz invariance manifest, handles gauge theories and constraints elegantly, links quantum theory directly to classical mechanics in the $ℏ \to 0$ limit, and provides the natural bridge to statistical mechanics through Wick rotation. The surprising fact: a classical concept—the action evaluated along arbitrary, even wildly non-differentiable, paths—encodes the entire quantum content of the theory, with the classical trajectory emerging only as the stationary-phase point of a violently oscillatory integral.

The Core Insight

Quantum amplitudes are obtained by superposing contributions from every conceivable history, not just the classical one. Each history $ϕ (x)$ contributes a complex number of unit modulus, $e^{i S [ϕ] /ℏ}$ . Constructive interference near the stationary point of $S$ reproduces classical physics; everywhere else, rapid phase oscillation produces cancellation. The transition from a single trajectory to a democratic sum over all of them is the conceptual heart of the formalism.

Formal Statement

Quantum-Mechanical Path Integral (Single Particle)

For a particle with Lagrangian $L (q, \overset{q}{˙})$ , the transition amplitude between position eigenstates is given by the configuration-space path integral:

⟨ q_{f}, t_{f} ∣ q_{i}, t_{i} ⟩ = \int_{q (t_{i}) = q_{i}}^{q (t_{f}) = q_{f}} D q (t) exp (\frac{i}{ℏ} \int_{t_{i}}^{t_{f}} L (q, \overset{q}{˙}) d t)

where $D q (t)$ denotes integration over all paths with the stated endpoints, $S [q] = \int L d t$ is the classical action, and $ℏ$ is the reduced Planck constant.

⟳Insert

N - 1

complete sets of position eigenstates between time-sliced evolution operators

e^{- i H ϵ /ℏ}

, with

ϵ = (t_{f} - t_{i}) / N

, evaluate each short-time kernel, and take

N \to \infty

The Time-Slicing Measure

The functional measure is the formal limit of a finite-dimensional integral over intermediate positions:

D q (t) = N \to \infty lim (\frac{m}{2 π i ℏ ϵ})^{N /2} k = 1 \prod N - 1 d q_{k}

where $m$ is the particle mass and the prefactor is the normalization of each Gaussian short-time kernel.

⟳The free-particle short-time kernel

⟨ q_{k + 1} ∣ e^{- i H ϵ /ℏ} ∣ q_{k} ⟩

is a Gaussian in

(q_{k + 1} - q_{k})

; its normalization

(m /2 π i ℏ ϵ)^{1/2}

accumulates over

N

slices.

Field-Theoretic Generating Functional

For a scalar field $ϕ (x)$ , all quantum information is packaged in the generating functional with external source $J (x)$ :

Z [J] = \int D ϕ exp (\frac{i}{ℏ} \int d^{4} x [L (ϕ, \partial_{μ} ϕ) + J (x) ϕ (x)])

where $L$ is the Lagrangian density and $J (x)$ is a classical source coupling linearly to the field.

⟳Promote the single coordinate

q (t)

to a field

ϕ (x, t)

indexed continuously by space, replacing

\int L d t

\int L d^{4} x

and the discrete product over

q_{k}

by a product over field values at every spacetime point.

Correlation Functions from Functional Derivatives

Time-ordered vacuum correlation functions (Green's functions) are generated by repeated functional differentiation:

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

where $T$ denotes time ordering and the division by $Z [0]$ removes vacuum bubble contributions.

⟳Each functional derivative

δ / δ J (x_{k})

brings down a factor

(i /ℏ) ϕ (x_{k})

from the exponential; setting

J = 0

leaves the

n

-point integrand, and the path integral automatically enforces time ordering.

Euclidean Path Integral and Wick Rotation

Analytically continuing to imaginary time $t \to - i τ$ converts the oscillatory weight into a real, positive, exponentially damped one:

Z_{E} [J] = \int D ϕ exp (- \frac{1}{ℏ} S_{E} [ϕ] + \frac{1}{ℏ} \int d^{4} x_{E} J ϕ), S_{E} = \int d^{4} x_{E} [\frac{1}{2} (\partial ϕ)^{2} + V (ϕ)]

where $S_{E} \geq 0$ is the Euclidean action and $d^{4} x_{E}$ is the Euclidean volume element.

⟳Substitute

t = - i τ

into the Minkowski action; the factor

i d t = d τ

and the sign flip in

(\partial_{t} ϕ)^{2}

turn

i S

into

- S_{E}

, rendering the measure a genuine probability weight.

Derivation of the Path Integral from Operator Mechanics

We derive Eq. [1] from canonical quantization in numbered steps.

Slice the evolution operator. Write $⟨ q_{f} ∣ e^{- i H (t_{f} - t_{i}) /ℏ} ∣ q_{i} ⟩$ and split the interval into $N$ equal pieces of width $ϵ$ , so $e^{- i H T /ℏ} = (e^{- i H ϵ /ℏ})^{N}$ .

Insert completeness. Between adjacent factors insert $\int d q_{k} ∣ q_{k} ⟩ ⟨ q_{k} ∣ = 1$ , yielding a product of short-time kernels $\prod_{k} ⟨ q_{k + 1} ∣ e^{- i H ϵ /ℏ} ∣ q_{k} ⟩$ .

Evaluate the kernel. For $H = p^{2} /2 m + V (q)$ , insert momentum completeness and perform the Gaussian $p$ -integral:

⟨ q_{k + 1} ∣ e^{- i H ϵ /ℏ} ∣ q_{k} ⟩ \approx (\frac{m}{2 π i ℏ ϵ})^{1/2} exp [\frac{i ϵ}{ℏ} (\frac{m}{2} (\frac{q _{k + 1} - q _{k}}{ϵ})^{2} - V (q_{k}))] .

⟳Complete the square in

p

after writing

e^{- iV ϵ /ℏ}

to the right; the

p

-integral is a shifted Gaussian.

Recognize the action. The exponent summed over slices is $\frac{i ϵ}{ℏ} \sum_{k} [\frac{m}{2} \overset{q}{˙}_{k}^{2} - V (q_{k})] \to \frac{i}{ℏ} \int L d t$ as $ϵ \to 0$ .

Take the continuum limit. The accumulated normalizations form the measure of Eq. [2], and the product of integrals becomes the functional integral of Eq. [1]. $■$

Richard P. Feynman | 1918–1988

Feynman, building on a 1933 paper by Dirac that noted the amplitude "corresponds to" $e^{i L d t /ℏ}$ , established the path integral as a full reformulation of quantum mechanics in his 1942 Princeton PhD thesis (supervised by Wheeler) and his landmark 1948 Reviews of Modern Physics paper. He used it to derive the Feynman rules and diagrams for QED, work for which he shared the 1965 Nobel Prize with Schwinger and Tomonaga. His view of physics as a "sum over histories" reshaped quantum field theory, statistical mechanics, and our intuition for quantum interference.

Building Intuition

Picture the famous double-slit experiment, then push it to its logical extreme. Add more screens with more slits; the particle's amplitude is the sum over each path through the holes. Now remove the screens entirely and drill "all possible holes everywhere"—the particle effectively traverses every path through space simultaneously, each contributing a phase $e^{i S /ℏ}$ . Far from the classical trajectory, neighboring paths have wildly different actions, so their phases point in random directions and cancel. Near the classical path, where $δ S = 0$ , neighboring paths share nearly the same phase and add coherently. Thus the classical trajectory is not "chosen" but emerges as the locus of constructive interference. In QFT, the "particle" becomes a field configuration over all spacetime, and the sum over histories becomes a sum over field shapes—a superposition of every way the field can wiggle between specified boundary data.

Stationary-phase: classical path dominates by constructive interference

<svg viewBox="0 0 460 280" xmlns="http://www.w3.org/2000/svg">
<rect width="460" height="280" fill="#0a0e17"/>
<text x="30" y="30" fill="#e2e8f0" font-size="13">Sum over paths q(t)</text>
<circle cx="60" cy="160" r="4" fill="#e2e8f0"/>
<text x="40" y="185" fill="#e2e8f0" font-size="11">q_i, t_i</text>
<circle cx="400" cy="120" r="4" fill="#e2e8f0"/>
<text x="380" y="110" fill="#e2e8f0" font-size="11">q_f, t_f</text>
<!-- non-classical wiggly paths -->
<path d="M60,160 C140,60 200,230 280,80 S360,200 400,120" stroke="#94a3b8" fill="none" stroke-width="1" opacity="0.5"/>
<path d="M60,160 C130,220 220,40 300,210 S370,70 400,120" stroke="#94a3b8" fill="none" stroke-width="1" opacity="0.5"/>
<path d="M60,160 C150,100 210,200 290,100 S350,180 400,120" stroke="#94a3b8" fill="none" stroke-width="1" opacity="0.4"/>
<!-- classical path -->
<path d="M60,160 C160,130 280,120 400,120" stroke="#d4a853" fill="none" stroke-width="2.5"/>
<text x="170" y="150" fill="#d4a853" font-size="12">classical: δS=0</text>
<text x="180" y="250" fill="#94a3b8" font-size="11">others cancel by oscillating phase e^{iS/ℏ}</text>
</svg>

Worked Examples

Free-Particle Propagator

Given: A free particle of mass

m

, with

V = 0

, propagating from

q_{i}

t_{i}

q_{f}

t_{f}

over duration

T = t_{f} - t_{i}

. Steps:

The classical action for the straight-line trajectory is $S_{cl} = \frac{m ( q _{f} - q _{i} ) ^{2}}{2 T}$ .
Because $L$ is quadratic, the path integral is exactly Gaussian; fluctuations contribute only a normalization $A (T)$ , giving $K = A (T) e^{i S_{cl} /ℏ}$ .
Determine $A (T)$ from the Gaussian fluctuation determinant (or from the semigroup property $\int d q^{'} K (q_{f}, T_{2}; q^{'}) K (q^{'}, T_{1}; q_{i}) = K (q_{f}, T_{1} + T_{2}; q_{i})$ ), yielding $A (T) = (m /2 π i ℏ T)^{1/2}$ .

Answer:

K (q_{f}, t_{f}; q_{i}, t_{i}) = (\frac{m}{2 π i ℏ T})^{1/2} exp [\frac{im ( q _{f} - q _{i} ) ^{2}}{2ℏ T}]

Numerically, for an electron ( $m = 9.11 \times 1 0^{- 31}$ kg), $Δ q = 1$ nm, $T = 1 0^{- 15}$ s: the phase $S_{cl} /ℏ = m Δ q^{2} / (2ℏ T) \approx (9.11 \times 1 0^{- 31} \cdot 1 0^{- 18}) / (2 \cdot 1.055 \times 1 0^{- 34} \cdot 1 0^{- 15}) \approx 4.3$ radians (dimensionless), confirming strong quantum interference at this scale.

Harmonic Oscillator via Fluctuation Determinant

Given: A harmonic oscillator

L = \frac{m}{2} \overset{q}{˙}^{2} - \frac{m ω ^{2}}{2} q^{2}

, duration

T

. Steps:

Split $q (t) = q_{cl} (t) + η (t)$ with $η$ vanishing at endpoints; the action separates as $S = S_{cl} + \frac{m}{2} \int (\overset{η}{˙}^{2} - ω^{2} η^{2}) d t$ .
The classical action is $S_{cl} = \frac{mω}{2 s i n ω T} [(q_{i}^{2} + q_{f}^{2}) cos ω T - 2 q_{i} q_{f}]$ .
The fluctuation integral gives $det^{- 1/2}$ of the operator $- \partial_{t}^{2} - ω^{2}$ ; its ratio to the free case is $ω T / sin ω T$ , fixing the prefactor.

Answer:

K = (\frac{mω}{2 π i ℏ sin ω T})^{1/2} exp [\frac{imω}{2ℏ sin ω T} ((q_{i}^{2} + q_{f}^{2}) cos ω T - 2 q_{i} q_{f})]

For $ω T \to 0$ this reduces smoothly to the free-particle result; poles at $ω T = nπ$ signal the focusing of all classical paths (caustics).

Common Misconceptions

The measure $D ϕ$ is not a genuine measure. In Minkowski signature the "weight"

e^{i S /ℏ}

has unit modulus and the formal infinite-dimensional integral does not converge absolutely; it is defined only through limiting/regularization procedures (time-slicing, Euclidean continuation, lattice). Treating

D ϕ e^{i S /ℏ}

as a literal Lebesgue-type measure is mathematically false—no translation-invariant countably-additive measure exists on infinite-dimensional spaces. Only the Euclidean version, Eq. [5], approaches a true (Wiener-like, Gaussian-based) measure, and even then interactions require renormalization. Naïve path integration violates operator ordering. One might think the configuration-space measure of Eq. [2] always captures the Hamiltonian faithfully. It does not when the kinetic term is position-dependent—e.g. a particle on a curved manifold with

H = \frac{1}{2} g^{ij} (q) p_{i} p_{j}

. The naïve discretization

g_{ij} (q_{k}) \overset{q}{˙}^{i} \overset{q}{˙}^{j}

misses an

O (ℏ^{2})

curvature term (proportional to the Ricci scalar

R

) that depends on the discretization point (midpoint vs. endpoint). The correct prescription comes from the phase-space path integral with careful ordering, demonstrating that the configuration-space form is not universally valid.

Connections

The path integral sits at the crossroads of much of theoretical physics. It reduces to ordinary Canonical Quantization via time-slicing, and the two are equivalent for well-behaved Hamiltonians. The Euclidean version, Eq. [5], is mathematically identical to the partition function of Statistical Mechanics, with $ℏ$ playing the role of temperature and the action playing the role of energy—this is the bedrock of Quantum Field Theory at Finite Temperature. It provides the natural language for Gauge Theory through the Faddeev–Popov procedure and ghost fields, and underlies the Renormalization Group when one integrates out high-momentum field modes shell by shell. The stationary-phase limit connects it directly to Classical Mechanics and the Hamilton-Jacobi Equation, while instanton solutions of Eq. [5] describe Quantum Tunneling non-perturbatively.

Real-World Applications

Quantum Electrodynamics (QED): Feynman diagrams and the precision computation of the electron anomalous magnetic moment $g - 2$ (agreement with experiment to 12 significant figures) derive directly from the path integral generating functional.
Quantum Chromodynamics (QCD) and Lattice Gauge Theory: The Euclidean path integral discretized on a spacetime lattice (Wilson, 1974) is the only systematic non-perturbative method to compute hadron masses and confinement, run on supercomputers worldwide.
Standard Model & Higgs Mechanism: Spontaneous symmetry breaking, gauge fixing, and electroweak loop corrections are all formulated and computed through path-integral effective potentials.
Condensed Matter Physics: Coherent-state and Hubbard–Stratonovich path integrals describe superconductivity (BCS), the quantum Hall effect, and quantum phase transitions.
Instantons & Tunneling Rates: False-vacuum decay (Coleman bounce), the rate of baryon-number violation via sphalerons, and the theta-vacuum structure of QCD are computed from saddle points of the Euclidean action.
Quantum Gravity & Cosmology: The Hartle–Hawking "no-boundary" wavefunction of the universe and Euclidean black-hole thermodynamics (Gibbons–Hawking) are defined by gravitational path integrals.
Stochastic Processes & Finance: The formal identity between Euclidean path integrals and the Feynman–Kac formula underlies Brownian-motion option pricing (Black–Scholes) and polymer physics.

Historical Development

1923–1925: Wiener constructs the rigorous measure for Brownian motion—the Euclidean ancestor of the path integral.
1933: Dirac, in Physikalische Zeitschrift der Sowjetunion, observes that the quantum transition amplitude over a short time "corresponds to" $exp (i L d t /ℏ)$ , planting the seed.
1942: Feynman's Princeton PhD thesis, "The Principle of Least Action in Quantum Mechanics," develops the sum-over-histories idea fully.
1948: Feynman publishes "Space-Time Approach to Non-Relativistic Quantum Mechanics" (Rev. Mod. Phys.), establishing the formalism; companion work derives QED Feynman rules.
1949: Dyson proves the equivalence of Feynman's, Schwinger's, and Tomonaga's formulations of QED.
1950s–60s: Matthews–Salam, and later Faddeev–Popov (1967), extend the path integral to non-abelian gauge theories, introducing gauge-fixing and ghosts.
1972: 't Hooft and Veltman use path-integral (dimensional) regularization to prove the renormalizability of gauge theories.
1974: Wilson formulates lattice gauge theory, making the Euclidean path integral a computational tool and explaining confinement.
1975–83: Coleman (instantons/false vacuum), Gibbons–Hawking and Hartle–Hawking apply Euclidean path integrals to tunneling, black-hole thermodynamics, and quantum cosmology.

Subtleties & Limits

Despite its practical triumphs, the path integral resists full mathematical rigor. The Minkowski measure $D ϕ e^{i S /ℏ}$ has no known countably-additive realization; constructive quantum field theory (Glimm–Jaffe) succeeds only for super-renormalizable models in $d \leq 3$ and remains open for four-dimensional Yang–Mills (a Clay Millennium Problem). The gravitational path integral faces the conformal-factor problem (the Euclidean Einstein–Hilbert action is unbounded below) and ambiguities in the contour of integration. Convergence and the sign problem plague Euclidean lattice simulations at finite chemical potential, where the weight becomes complex. Finally, the precise definition of the functional measure including its anomalous Jacobians (Fujikawa's derivation of chiral anomalies) reveals that the "measure" itself carries physical content not visible in the classical action—an aspect still being mapped in curved space and on the lattice.