Path Integral Formulation

The path integral formulation of quantum mechanics expresses transition amplitudes as a sum over all possible paths connecting initial and final configurations, each weighted by a phase $e^{i S /ℏ}$ where $S$ is the classical action. Formulated by Richard Feynman (1948), it is equivalent to canonical (operator) quantum mechanics but reorganizes the theory around the Lagrangian and trajectories rather than Hamiltonians and operators. It is the foundation of modern quantum field theory, statistical mechanics, and many areas of mathematical physics.

The path integral formulation states that the quantum-mechanical amplitude for a system to evolve from configuration $q_{a}$ at time $t_{a}$ to $q_{b}$ at time $t_{b}$ is obtained by integrating $e^{i S [q] /ℏ}$ over the infinite-dimensional space of all paths $q (t)$ satisfying the boundary conditions. This reorganization matters because it makes symmetries, the classical limit, and gauge structure transparent, and because it generalizes seamlessly to fields, providing the dominant computational and conceptual framework of quantum field theory. The surprising fact is that the non-differentiable, "zigzag" paths — which have infinite kinetic energy classically — dominate the integral, yet the smooth classical path emerges as the stationary-phase contribution in the $ℏ \to 0$ limit. The "measure" $D q$ is not a genuine countably-additive measure in the Lebesgue sense, making the formulation mathematically subtle but physically extraordinarily robust.

The Core Insight

Classical mechanics selects the single path of stationary action via $δ S = 0$ . Quantum mechanics keeps all paths but assigns each a complex phase of unit modulus, $e^{i S /ℏ}$ . Constructive interference near the classical trajectory (where $S$ is stationary, so phases align) reproduces classical behavior; destructive interference elsewhere suppresses non-classical paths. The amplitude is the coherent superposition of every conceivable history.

Formal Statement

Propagator / Kernel

The central object is the propagator (or kernel), the position-space matrix element of the time-evolution operator:

K (q_{b}, t_{b}; q_{a}, t_{a}) = ⟨ q_{b} ∣ e^{- i \hat{H} (t_{b} - t_{a}) /ℏ} ∣ q_{a} ⟩

⟳Insert the time-evolution operator

\hat{U} (t_{b}, t_{a}) = e^{- i \hat{H} (t_{b} - t_{a}) /ℏ}

(for time-independent

\hat{H}

) between position eigenstates

\overset{q}{^} ∣ q ⟩ = q ∣ q ⟩

Here $\hat{H}$ is the Hamiltonian operator, $∣ q ⟩$ are position eigenstates with $⟨ q^{'} ∣ q ⟩ = δ (q^{'} - q)$ , and $K$ is the amplitude that propagates the wavefunction: $ψ (q_{b}, t_{b}) = \int d q_{a} K (q_{b}, t_{b}; q_{a}, t_{a}) ψ (q_{a}, t_{a})$ .

Feynman Path Integral

The propagator is expressed as a functional integral over all paths $q (t)$ with $q (t_{a}) = q_{a}$ , $q (t_{b}) = q_{b}$ :

K (q_{b}, t_{b}; q_{a}, t_{a}) = \int_{q (t_{a}) = q_{a}}^{q (t_{b}) = q_{b}} D q (t) exp (\frac{i}{ℏ} S [q])

⟳Obtained by time-slicing the propagator into

N

infinitesimal evolutions and taking

N \to \infty

(see proof below).

where the action functional is the time integral of the Lagrangian:

S [q] = \int_{t_{a}}^{t_{b}} d t L (q (t), \overset{q}{˙} (t), t), L = \frac{1}{2} m \overset{q}{˙}^{2} - V (q)

⟳

L

is the classical Lagrangian; for a non-relativistic particle in potential

V

, kinetic minus potential energy.

The symbol $D q (t)$ denotes the (formal) integration measure over path space, defined as a limit of finite-dimensional integrals.

Time-Slicing Definition of the Measure

The path-integral measure is rigorously defined (in the discretized sense) by

\int D q e^{i S /ℏ} = N \to \infty lim (\frac{m}{2 π i ℏ ϵ})^{N /2} \int k = 1 \prod N - 1 d q_{k} exp (\frac{i}{ℏ} k = 0 \sum N - 1 [\frac{m ( q _{k + 1} - q _{k} ) ^{2}}{2 ϵ} - ϵ V (q_{k})])

⟳Split

[t_{a}, t_{b}]

into

N

slices of width

ϵ = (t_{b} - t_{a}) / N

; insert

N - 1

completeness relations

\int d q_{k} ∣ q_{k} ⟩ ⟨ q_{k} ∣ = 1

; evaluate each short-time kernel.

with $q_{0} = q_{a}$ , $q_{N} = q_{b}$ , $ϵ = (t_{b} - t_{a}) / N$ , and the normalization factor $(m /2 π i ℏ ϵ)^{N /2}$ absorbed into $D q$ .

Why It Holds / Derivation

Time-Slicing Derivation

We derive [2]–[4] from [1] for $\hat{H} = \overset{p}{^}^{2} /2 m + V (\overset{q}{^})$ .

Step 1 — Slice the evolution. Write

t_{b} - t_{a} = N ϵ

and split the exponential:

e^{- i \hat{H} (t_{b} - t_{a}) /ℏ} = (e^{- i \hat{H} ϵ /ℏ})^{N}

⟳The operator exponential factorizes because

\hat{H}

commutes with itself;

N

identical factors.

Step 2 — Insert completeness. Between each factor insert

\int d q_{k} ∣ q_{k} ⟩ ⟨ q_{k} ∣ = 1

, giving

K = \int k = 1 \prod N - 1 d q_{k} k = 0 \prod N - 1 ⟨ q_{k + 1} ∣ e^{- i \hat{H} ϵ /ℏ} ∣ q_{k} ⟩

⟳

N - 1

resolutions of identity produce

N - 1

integrals and a product of

N

short-time matrix elements.

Step 3 — Trotter factorization. For small

ϵ

, the Baker–Campbell–Hausdorff error in splitting

e^{- i ϵ (\hat{T} + \hat{V}) /ℏ} \approx e^{- i ϵ \hat{T} /ℏ} e^{- i ϵ \hat{V} /ℏ}

O (ϵ^{2})

⟨ q_{k + 1} ∣ e^{- i \hat{H} ϵ /ℏ} ∣ q_{k} ⟩ \approx ⟨ q_{k + 1} ∣ e^{- i ϵ \overset{p}{^}^{2} /2 m ℏ} ∣ q_{k} ⟩ e^{- i ϵ V (q_{k}) /ℏ}

⟳

\hat{V} (\overset{q}{^})

is diagonal in position; act it on

∣ q_{k} ⟩

to get the c-number

V (q_{k})

. The Trotter product formula guarantees convergence as

N \to \infty

Step 4 — Evaluate the kinetic factor. Insert momentum completeness

\int \frac{d p}{2 π ℏ} ∣ p ⟩ ⟨ p ∣

and use

⟨ q ∣ p ⟩ = e^{i pq /ℏ}

⟨ q_{k + 1} ∣ e^{- i ϵ \overset{p}{^}^{2} /2 m ℏ} ∣ q_{k} ⟩ = \int \frac{d p _{k}}{2 π ℏ} exp [\frac{i}{ℏ} (p_{k} (q_{k + 1} - q_{k}) - \frac{ϵ p _{k}^{2}}{2 m})]

⟳A Gaussian integral in

p_{k}

; complete the square in the exponent.

Step 5 — Do the Gaussian momentum integral.

\int \frac{d p _{k}}{2 π ℏ} e^{\frac{i}{ℏ} (p_{k} Δ q_{k} - ϵ p_{k}^{2} /2 m)} = (\frac{m}{2 π i ℏ ϵ})^{1/2} exp [\frac{im ( Δ q _{k} ) ^{2}}{2ℏ ϵ}]

⟳Standard Fresnel/Gaussian integral

\int d p e^{- a p^{2} + b p} = π / a e^{b^{2} /4 a}

with imaginary

a

; here

Δ q_{k} = q_{k + 1} - q_{k}

Step 6 — Assemble. Combining the

N

factors and recognizing the Riemann sum:

k = 0 \sum N - 1 [\frac{m}{2} (\frac{Δ q _{k}}{ϵ})^{2} - V (q_{k})] ϵ N \to \infty \int_{t_{a}}^{t_{b}} d t L = S [q]

⟳

Δ q_{k} / ϵ \to \overset{q}{˙}

\sum_{k} ϵ \to \int d t

; the exponent becomes

i S /ℏ

, completing [2] and [4].

■

The "limit" in [4] is not the limit of integrals against a genuine measure. The factor $(m /2 π i ℏ ϵ)^{N /2}$ diverges as $N \to \infty$ , and the oscillatory integrand is not absolutely convergent. The path integral exists as a limit of finite-dimensional oscillatory integrals (or via analytic continuation from the Euclidean theory), not as a Lebesgue integral over path space. There is no translation-invariant $σ$ -additive measure on the infinite-dimensional path space — see Wiener Measure for the rigorous Euclidean counterpart.

Richard P. Feynman | 1918–1988

Feynman developed the path integral in his 1942 Princeton PhD thesis under John Wheeler and published the definitive account in Reviews of Modern Physics (1948), building on a cryptic remark by Dirac (1933) that the short-time propagator "corresponds to" $e^{i L ϵ /ℏ}$ . Feynman turned "corresponds to" into "equals" (up to normalization), reformulating all of quantum mechanics as a sum over histories. He later extended it to electrodynamics, producing the Feynman diagrams (1949) that organize perturbation theory. He shared the 1965 Nobel Prize with Schwinger and Tomonaga. Feynman's path integral remains his deepest single contribution to theoretical physics, prized for making quantum mechanics visualizable as interfering histories.

Building Intuition

Imagine sending the particle from $A$ to $B$ and asking it to "try" every conceivable route — straight lines, loops, paths that briefly run backward, infinitely jagged zigzags. Each route carries a stopwatch hand (a unit complex phasor) that rotates at rate $S /ℏ$ . To get the total amplitude, add all the little arrows tip-to-tail. Where $S$ changes rapidly between neighboring paths (far from the classical trajectory), the arrows spin wildly and cancel. Near the classical path, $δ S = 0$ means neighboring paths have nearly equal $S$ , so their arrows point the same way and add coherently. This is why classical mechanics ( $ℏ \to 0$ ) emerges: the classical path is where the phasors stop canceling. The width of the "corridor" of contributing paths is set by $ℏ$ — vanishingly thin for macroscopic objects, broad for electrons.

Path interference: the classical trajectory is the stationary-phase corridor among all summed paths

Worked Examples

Free Particle Propagator

Given: A free particle,

V = 0

, mass

m

, propagating from

q_{a}

t_{a} = 0

q_{b}

at time

T

. Step 1 — Classical action. The classical path is the straight line

q_{c l} (t) = q_{a} + (q_{b} - q_{a}) t / T

with constant velocity

v = (q_{b} - q_{a}) / T

. Then

S_{c l} = \int_{0}^{T} \frac{1}{2} m v^{2} d t = \frac{m ( q _{b} - q _{a} ) ^{2}}{2 T}

⟳Constant velocity gives constant Lagrangian; multiply by

T

Step 2 — Fluctuation integral. Write

q = q_{c l} + η

with

η (0) = η (T) = 0

. Since

S

is quadratic,

S = S_{c l} + \frac{1}{2} \int \overset{η}{˙}^{2} m d t

and the fluctuation determinant is field-independent, yielding the prefactor

(m /2 π i ℏ T)^{1/2}

(the continuum limit of the slicing normalization). Step 3 — Result.

K (q_{b}, T; q_{a}, 0) = (\frac{m}{2 π i ℏ T})^{1/2} exp [\frac{im ( q _{b} - q _{a} ) ^{2}}{2ℏ T}]

⟳Combine

S_{c l}

from [11] with the Gaussian fluctuation prefactor; verify it satisfies the free Schrödinger equation and reduces to

δ (q_{b} - q_{a})

T \to 0

Numerical check (units). For an electron (

m = 9.11 \times 1 0^{- 31}

kg),

q_{b} - q_{a} = 1

= 1 0^{- 9}

T = 1 0^{- 15}

ℏ = 1.055 \times 1 0^{- 34}

J·s: the phase

= \frac{m ( q _{b} - q _{a} ) ^{2}}{2ℏ T} = \frac{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 \times 1 0^{0} rad \approx 4.3 rad

(dimensionless).

Harmonic Oscillator Propagator

Given:

L = \frac{1}{2} m \overset{q}{˙}^{2} - \frac{1}{2} m ω^{2} q^{2}

, propagation over time

T

(with

ω T \neq = nπ

). Step 1 — Classical action. Solving

\overset{q}{¨}_{c l} + ω^{2} q_{c l} = 0

with boundary values and substituting gives the well-known result:

S_{c l} = \frac{mω}{2 sin ω T} [(q_{a}^{2} + q_{b}^{2}) cos ω T - 2 q_{a} q_{b}]

⟳Use

q_{c l} (t) = A cos ω t + B sin ω t

, integrate by parts so

S_{c l} = \frac{m}{2} [q \overset{q}{˙}]_{0}^{T}

, evaluate boundary terms.

Step 2 — Fluctuation determinant. The quadratic fluctuation operator

- m (\partial_{t}^{2} + ω^{2})

gives, via the Gelfand–Yaglom theorem, the prefactor

(mω /2 π i ℏ sin ω T)^{1/2}

. Step 3 — Result.

K = (\frac{mω}{2 π i ℏ sin ω T})^{1/2} exp (\frac{i}{ℏ} S_{c l})

⟳Reduces to the free result [12] as

ω \to 0

(

sin ω T \to ω T

); poles at

ω T = nπ

encode the energy spectrum via the Mehler/spectral expansion.

Numerical check. Take

ω = 1 0^{14}

s⁻¹,

T = π / (4 ω) = 7.85 \times 1 0^{- 15}

q_{a} = q_{b} = 0

. Then

S_{c l} = 0

, and the amplitude magnitude is

∣ K ∣ = (mω /2 π ℏ sin ω T)^{1/2}

. With

m = 9.11 \times 1 0^{- 31}

kg:

mω / (2 π ℏ sin (π /4)) = (9.11 \times 1 0^{- 17}) / (2 π \cdot 1.055 \times 1 0^{- 34} \cdot 0.707) = 1.94 \times 1 0^{17} m^{- 1}

, so

∣ K ∣ \approx 4.4 \times 1 0^{8} m^{- 1/2}

Common Misconceptions

The "measure" is not a measure. Students often treat

D q

like a Lebesgue measure and manipulate the path integral with the same freedom as a finite integral. In Minkowski (real-time) signature,

e^{i S /ℏ}

has unit modulus everywhere, so the integral does not converge absolutely and no genuine measure exists. Rigorous statements require either the discretized limit [4], analytic continuation to imaginary time (Euclidean Wiener Measure), or oscillatory-integral (Fresnel) machinery. Formal manipulations frequently give correct answers, but their validity is non-trivial. Operator ordering hides in the measure. The naive replacement

\hat{H} \to H (q, p)

is ambiguous when

\overset{q}{^}

and

\overset{p}{^}

do not commute (e.g. for position-dependent mass or magnetic terms). Different discretizations (midpoint vs endpoint evaluation of

V

or of velocity-dependent terms) correspond to different operator orderings (Weyl, normal, etc.). The midpoint prescription corresponds to Weyl ordering. A "Lagrangian" path integral silently encodes an ordering choice. Naively integrating the kinetic phase gives the wrong normalization without the prefactor. Suppose one writes

K = \int D q e^{i S /ℏ}

and tries to evaluate the free case without tracking the

(m /2 π i ℏ ϵ)^{N /2}

normalization. One then obtains a propagator missing the crucial

(m /2 π i ℏ T)^{1/2}

factor, so

K

fails the composition law

\int d q_{c} K (b, c) K (c, a) = K (b, a)

and does not reduce to

δ (q_{b} - q_{a})

T \to 0

. The normalization is not optional decoration — it is fixed by unitarity and is part of the definition of

D q

Connections

The path integral connects to the rest of physics along several axes. Via Wick Rotation $t \to - i τ$ , the oscillatory factor $e^{i S /ℏ}$ becomes the convergent Boltzmann-like weight $e^{- S_{E} /ℏ}$ , linking quantum mechanics to Statistical Mechanics: the Euclidean partition function $Z = \int D q e^{- S_{E} /ℏ}$ over periodic paths is the thermal partition function with $β ℏ =$ period. Generalizing the coordinate $q (t)$ to a field $ϕ (x, t)$ yields the Quantum Field Theory path integral, the workhorse of particle physics. The stationary-phase ( $ℏ \to 0$ ) limit reproduces Classical Mechanics through the Principle of Least Action and connects to WKB Approximation and semiclassical methods. The operator formulation is recovered through the Canonical Quantization correspondence, and for gauge theories the path integral requires Faddeev–Popov Gauge Fixing.

Real-World Applications

Quantum Field Theory & the Standard Model. Cross sections in the LHC are computed from $S$ -matrix elements derived from the QFT path integral via Feynman diagrams — the perturbative expansion of $\int D ϕ e^{i S /ℏ}$ .
Lattice QCD. The Euclidean path integral for the strong interaction is evaluated numerically by Monte Carlo importance sampling on a spacetime lattice, predicting hadron masses (proton mass to ~1%) from first principles.
Polymer physics & critical phenomena. The Euclidean path integral maps directly onto the statistics of polymer conformations and to the partition functions of statistical field theory near second-order phase transitions (Wilson's renormalization group).
Condensed matter / Berry phase & topology. Coherent-state and instanton path integrals explain the quantum Hall effect, superconducting tunneling (Josephson junctions), and topological terms (theta vacua, Wess–Zumino).
Quantum tunneling & decay rates. Instantons — finite-action solutions of the Euclidean equations — give exponentially small tunneling amplitudes $e^{- S_{E} /ℏ}$ for false-vacuum decay and chemical reaction rates.
Financial mathematics. The Feynman–Kac formula recasts option-pricing PDEs (Black–Scholes) as path integrals / expectations over stochastic price paths, a direct descendant of the Euclidean construction.
Quantum computing & open systems. Influence functionals (Feynman–Vernon) describe decoherence of a system coupled to an environment by integrating out bath paths.

Historical Development

1933 — P. A. M. Dirac publishes "The Lagrangian in Quantum Mechanics," noting the short-time propagator corresponds to $exp (i L d t /ℏ)$ , planting the seed.
1942 — Richard Feynman, in his Princeton PhD thesis under John Wheeler, develops a space-time formulation of nonrelativistic quantum mechanics as a sum over paths.
1948 — Feynman publishes "Space-Time Approach to Non-Relativistic Quantum Mechanics" (Rev. Mod. Phys. 20, 367), the foundational paper of the path integral.
1948–49 — Mark Kac formalizes the Euclidean case via the Feynman–Kac formula, connecting the imaginary-time path integral to Brownian motion and Wiener measure.
1949 — Feynman introduces Feynman diagrams ("Space-Time Approach to Quantum Electrodynamics"), giving graphical rules for the perturbative path integral.
1950s — Extension to fields; Feynman–Vernon influence functional (1963) for open systems.
1965 — Feynman & Hibbs publish Quantum Mechanics and Path Integrals, the canonical textbook; Feynman shares the Nobel Prize.
1967 — Faddeev & Popov solve gauge-fixing for non-abelian path integrals, enabling the quantization of Yang–Mills theory.
1970s — Wilson, Polyakov, 't Hooft develop Euclidean lattice and instanton methods; the path integral becomes the central language of QFT and statistical field theory.

Subtleties & Limits

Several foundational and technical issues remain active:

Mathematical rigor in Minkowski signature. There is still no fully satisfactory, universally applicable definition of the real-time oscillatory path integral as a genuine integration theory; constructions rely on analytic continuation from Euclidean space, which can fail for theories with sign problems or no Osterwalder–Schrader reflection positivity.
The fermionic sign / sign problem. Monte Carlo evaluation of Euclidean path integrals at finite density or with real-time dynamics is obstructed by oscillatory/negative weights — an NP-hard problem in general, blocking ab initio simulation of dense QCD and many strongly correlated systems.
Measure for gauge and gravity. Defining $D g_{μν}$ for the gravitational path integral is unresolved; the Euclidean action is unbounded below (conformal factor problem), and the very meaning of "summing over geometries" (including topology change) is open.
Constructive QFT. Rigorously constructing interacting 4-dimensional path integrals (e.g. nontrivial $ϕ_{4}^{4}$ , or Yang–Mills with a mass gap — a Clay Millennium Problem) remains unsolved.
Operator ordering & curved configuration spaces. The path integral on a curved manifold requires extra $ℏ^{2}$ counterterms whose universal form is still debated.