Case Study 31.1: Feynman's Path Integral — A Revolutionary Perspective

How a Princeton graduate student overheard a remark by Dirac, took it literally, and reinvented quantum mechanics

The Puzzle

By 1941, quantum mechanics was twenty years old and spectacularly successful. The Schrödinger equation, Heisenberg's matrix mechanics, and Dirac's transformation theory had been synthesized into a coherent mathematical framework. The hydrogen atom was solved. Perturbation theory worked. Spin, identical particles, and the beginnings of quantum electrodynamics were all in place.

And yet there was a problem — specifically, a problem with quantum electrodynamics (QED). When physicists tried to compute higher-order corrections to processes like electron-photon scattering, they got infinity. Not metaphorical infinity — actual divergent integrals that could not be made to give finite answers. The framework that worked so beautifully for single-particle quantum mechanics seemed to break down when applied to quantum fields.

Richard Feynman, then a graduate student at Princeton working under John Archibald Wheeler, was obsessed with this problem. He believed that the standard formulation of quantum mechanics — based on Hamiltonians, operators, and the canonical commutation relations — might be part of the difficulty. The Hamiltonian formulation requires choosing a time direction, which sits uneasily with special relativity. Was there another way to formulate quantum mechanics, one that treated space and time more symmetrically?

The Dirac Connection

The seed of Feynman's idea came from a paper by Paul Dirac, published in 1933 in a Soviet physics journal (Physikalische Zeitschrift der Sowjetunion). In that paper, Dirac had noted an "analogy" between the quantum mechanical propagator and the exponential of the classical Lagrangian. Specifically, Dirac wrote that the short-time propagator $\langle x_{j+1}|e^{-i\hat{H}\epsilon/\hbar}|x_j\rangle$ "corresponds to" the expression $\exp(iL\epsilon/\hbar)$ where $L$ is the classical Lagrangian.

Dirac used the word "corresponds to" — an analogy, a suggestive parallel. He did not say "equals." He did not develop the idea further. It was a remark in passing, buried in a paper on other topics.

The story of what happened next has become part of physics folklore. In 1941, at a beer party at a Princeton tavern, Herbert Jehle (a visiting European physicist) mentioned Dirac's paper to Feynman. The next day, Feynman and Jehle went to the Princeton library, found the paper, and Feynman asked the critical question: "Did Dirac mean that they were equal, or merely analogous?"

Jehle did not know. Feynman decided to find out by computing. He worked through the short-time propagator for a simple system, discovered that Dirac's "analogy" was in fact an equality (up to a normalization constant), and then asked: what if you string together $N$ short-time propagators and take $N \to \infty$?

The answer was the path integral.

The Development

Feynman's Ph.D. thesis, completed in 1942 and supervised by Wheeler, was titled "The Principle of Least Action in Quantum Mechanics." In it, he developed the path integral formulation for the first time. The thesis was not published as a journal article until much later — World War II intervened, and Feynman spent the next three years at Los Alamos working on the Manhattan Project.

After the war, Feynman returned to the problem of QED with a new tool. The path integral formulation naturally suggested a perturbative expansion that could be represented graphically — what we now call Feynman diagrams. Each diagram corresponds to a term in the perturbative expansion of the path integral, and the "Feynman rules" for translating diagrams into mathematical expressions are a direct consequence of Wick's theorem applied to the Gaussian path integral.

In 1948, Feynman published his landmark paper "Space-Time Approach to Non-Relativistic Quantum Mechanics" in Reviews of Modern Physics. This paper presented the path integral for non-relativistic quantum mechanics — the content of this chapter. His subsequent papers (1949-1950) applied the path integral to QED, introducing the diagrams and renormalization techniques that tamed the infinities.

For this work, Feynman shared the 1965 Nobel Prize in Physics with Julian Schwinger and Sin-Itiro Tomonaga, who had independently developed equivalent (but operationally very different) approaches to renormalizing QED.

The Physics: Why It Matters

A New Foundation for the Classical Limit

In the Schrödinger formulation, the connection between quantum and classical mechanics requires Ehrenfest's theorem and a careful analysis of when expectation values track classical trajectories (which they sometimes do not — for example, in systems with chaotic classical limits).

In the path integral formulation, the classical limit is immediate and transparent. When $S_{\text{cl}} \gg \hbar$ (the classical action is much larger than Planck's constant), the stationary phase approximation selects the classical trajectory automatically. The principle of least action — the foundation of all classical mechanics — emerges as a consequence of constructive interference. There is no need for Ehrenfest's theorem, no need to worry about narrow wave packets, no need for any additional argument.

This alone would justify the path integral formulation. But there is much more.

Manifest Lorentz Invariance

The Hamiltonian formulation of quantum mechanics treats time differently from space: the Hamiltonian generates translations in time, while spatial translations are generated by the momentum operator. This asymmetry is deeply problematic for relativistic theories, where space and time should be treated on equal footing.

The Lagrangian that appears in the path integral, by contrast, is a Lorentz scalar. The action $S = \int L\, dt$ is Lorentz-invariant. The path integral $\int \mathcal{D}[\text{paths}]\, e^{iS/\hbar}$ therefore has manifest Lorentz invariance — no need to check at the end that your theory is relativistic. This is why the path integral became the standard formulation of quantum field theory.

Nonperturbative Physics

The path integral gives access to physical phenomena that are completely invisible in perturbation theory. Quantum tunneling, for example, is mediated by instantons — classical solutions of the Euclidean equations of motion. The tunneling amplitude goes as $e^{-S_E/\hbar}$, which is zero to all orders of perturbation theory in any coupling constant. Only the path integral, with its ability to identify and evaluate contributions from nontrivial saddle points, captures these effects.

Similarly, topological effects (like the Aharonov-Bohm phase) are naturally incorporated into the path integral through the topology of the space of paths. Paths that wind around a solenoid a different number of times contribute different phases, and summing over winding numbers produces the interference pattern.

Lattice Gauge Theory

The Wick-rotated (Euclidean) path integral is the foundation of lattice gauge theory, the primary tool for computing properties of the strong nuclear force (QCD) from first principles. The Euclidean path integral, with its real exponential weight $e^{-S_E/\hbar}$, can be evaluated numerically using Markov chain Monte Carlo methods — importance sampling the space of field configurations weighted by $e^{-S_E}$.

This approach has been spectacularly successful. Lattice QCD calculations have predicted hadron masses, computed the strong coupling constant, and provided the only rigorous evidence that QCD confines quarks. All of this rests on Feynman's path integral.

The Objections and Responses

Mathematical Rigor

The real-time path integral $\int \mathcal{D}[x]\, e^{iS/\hbar}$ is not a well-defined mathematical object in the sense of Lebesgue measure theory. The "measure" $\mathcal{D}[x]$ on the space of paths has never been rigorously constructed for the oscillatory case.

The Euclidean path integral is much better behaved: it corresponds to the Wiener measure on Brownian paths, which is rigorously constructed and well-studied in probability theory. The physicist's approach is to define the path integral as the limit of the discretized version and check that this limit gives the correct physics (which it does, in every case that has been checked).

Mathematicians continue to work on making the real-time path integral rigorous. This is an active and deep area of research at the intersection of functional analysis, probability theory, and mathematical physics. But for the working physicist, the discretized definition has never led to a wrong answer.

"It's Just the Schrödinger Equation in Disguise"

Some have argued that the path integral is merely a rewriting of the Schrödinger equation, with no genuinely new content. This is technically true — the path integral is exactly equivalent to the operator formulation for the systems considered in this course.

But this misses the point. Lagrangian mechanics is also "just" Newtonian mechanics in disguise, yet the Lagrangian formulation reveals symmetries (Noether's theorem), handles constraints naturally (generalized coordinates), and extends to field theory in ways that the Newtonian formulation cannot. Similarly, the path integral reveals the classical limit transparently, extends naturally to field theory, provides nonperturbative tools (instantons, lattice), and suggests computational methods (Feynman diagrams) that would be extremely difficult to discover from the operator formulation.

Different formulations of the same theory are not redundant. They are complementary windows onto the physics, and the deepest understanding comes from mastering all of them.

The Legacy

Feynman's path integral is arguably the single most influential idea in theoretical physics of the second half of the twentieth century. Its descendants include:

• Feynman diagrams: The universal language of particle physics
• Lattice gauge theory: The computational backbone of QCD
• The renormalization group: Wilson's reformulation using the path integral earned the 1982 Nobel Prize
• String theory: The worldsheet path integral (Polyakov action) is the starting point for perturbative string theory
• Quantum gravity approaches: Both loop quantum gravity and the gravitational path integral are descendants of Feynman's idea
• Condensed matter physics: Path integrals are the natural framework for studying phase transitions, superfluidity, superconductivity, and topological phases
• Stochastic quantization: Treating quantum field theory as a stochastic process, using the Euclidean path integral as a probability weight

It is not an exaggeration to say that the path integral is the common language of modern theoretical physics. Every theorist, regardless of subfield, uses it.

Reflection Questions

1. Feynman took Dirac's "analogy" literally and asked whether it was an equality. What does this tell us about the role of mathematical rigor versus physical intuition in theoretical physics?

2. The path integral and the Schrödinger equation give identical predictions. What criteria should we use to judge whether one formulation is "better" than another?

3. The real-time path integral is not mathematically rigorous, yet it has never given a wrong physical answer. Should physicists be troubled by this? How does this compare to other situations where formal mathematical manipulation outpaces rigorous justification (e.g., the early development of calculus)?

4. The Wick rotation connects quantum mechanics (oscillatory phases) to statistical mechanics (real exponentials). Is this a deep physical connection or a mathematical coincidence? What would it mean if the connection broke down for some system?

5. Feynman developed the path integral partly because he wanted a formulation that would extend naturally to relativistic quantum field theory. To what extent should the requirement of future generalizability guide the way we formulate current theories?