The Feynman Propagator and Causality

Relativistic quantum mechanics hands you two apparent disasters: solutions with negative energy, and a "probability" that can go negative. The Feynman propagator is the object that converts both liabilities into the working machinery of quantum field theory. It is a Green's function for the wave operator — the amplitude for a disturbance created at one spacetime point to be registered at another — but the wave equation alone does not fix which way time flows in that amplitude. That freedom is exactly the freedom to route a contour around two poles, and the Feynman routing, written universally as an infinitesimal $+i\epsilon$ , is the unique one in which positive-energy quanta travel forward in time while negative-energy modes travel backward and are relabelled as antiparticles moving forward. The deepest lesson here is a correction to a slogan almost everyone learns wrong first. Causality is not the statement that the propagator vanishes outside the light cone — it does not; it is exponentially small but nonzero for spacelike separations. What vanishes at spacelike separation is the field commutator $[\hat\phi(x),\hat\phi(y)]$ , and since every measurable quantity is built from commutators, no measurement outraces light even while virtual quanta appear to leak past the cone. This is microcausality, and it is the hinge on which the spin-statistics theorem turns. You arrive holding the nonrelativistic kernel of The Propagator from Path Integrals and the Wick rotation of Imaginary-Time and Euclidean Path Integrals; the pole at $p^2=m^2$ is the mass shell of The Klein-Gordon Equation. Here we build the causal propagator, prove by residues that the $+i\epsilon$ enforces the forward/backward split, and show that the invariant commutator function cancels to exactly zero for a spacelike interval. That single cancellation is why the field must be quantized with the right statistics — the necessity taken up in the track's closing article.

Key takeaways

The Feynman propagator $\Delta_F(x-y)=\langle 0|\,T\,\hat\phi(x)\hat\phi(y)\,|0\rangle$ is the Green's function of the Klein-Gordon operator, $(\Box_x+m^2)\Delta_F(x-y)=-\delta^4(x-y)$ , made unique by a $+i\epsilon$ prescription.
In momentum space the scalar Feynman propagator is $\tilde{\Delta}_F(p)=\frac{i}{p^2-m^2+i\epsilon}$ , while for a Dirac field the propagator is $\tilde{S}_F(p)=\frac{i(\slashed p+m)}{p^2-m^2+i\epsilon}$ . These are the propagator factors associated with scalar and fermion internal lines, respectively, in Feynman diagrams.
The $+i\epsilon$ pushes the $+\omega_p$ pole below the real axis and the $-\omega_p$ pole above it; closing the $p^0$ contour by residues gives positive-energy propagation forward in time and antiparticle propagation backward — the Stückelberg–Feynman picture.
Causality does not mean $\Delta_F$ vanishes outside the light cone (it does not); it means the commutator $[\hat\phi(x),\hat\phi(y)]$ vanishes for spacelike $(x-y)^2<0$ . Microcausality, not propagator support, is the causal statement.
The invariant commutator function $\Delta(x)$ cancels to exactly zero at spacelike separation because the two mode integrals map into each other under the Lorentz-invariant measure — the machine-checkable heart of causality.
Undoing the relativistic structure ( $E=p^2/2m$ ) recovers the nonrelativistic kernel $K$ of The Propagator from Path Integrals; the $+i\epsilon$ is the analytic continuation partner of the Wick rotation.
Key idea: inverting $p^2-m^2$ demands a contour choice, and the Feynman $+i\epsilon$ is the unique one that sends particles forward and antiparticles backward in time — a Lorentz-invariant Green's function whose commutator vanishes outside the light cone.

The core insight

Inverting $(p^2-m^2)$ requires a contour choice; the Feynman $+i\epsilon$ is the unique prescription that propagates positive-energy particles forward and negative-energy antiparticles backward in time, producing a Lorentz-invariant causal Green's function whose associated field commutator vanishes for spacelike separations.

Where this lives today

Every internal line in every Feynman diagram — from electron–muon scattering computed to ten digits at electron $g-2$ experiments, to Higgs-production amplitudes at the LHC — is a Feynman propagator carrying the $+i\epsilon$ . Automated amplitude generators such as the tools behind modern collider phenomenology are, at their core, engines for contracting fields into propagators and integrating over the loop momenta they carry. The same $+i\epsilon$ analytic structure determines the pole prescription used in perturbative amplitudes and underlies the analytic structure of scattering amplitudes, including the branch cuts associated with unitarity. Microcausality — the vanishing of the field commutator at spacelike separation — is likewise a fundamental structural requirement of relativistic QFT and plays a central role in dispersion relations, analyticity, and causal perturbation theory.

The physical picture

Picture the propagator as the amplitude for a ripple made at spacetime point $y$ to be found at $x$ . The Klein-Gordon operator $\Box+m^2$ tells you the ripple obeys the wave equation, but a differential operator has a kernel — you can add any solution of the homogeneous equation — so inverting it to get a Green's function is ambiguous. That ambiguity is not a nuisance; it is where all the physics of time's arrow lives. In momentum space the ambiguity is concrete: the energy integral $e^{-ip^0t}/(p^2-m^2)$ has poles on the real axis at $p^0=\pm\omega_{\bf p}$ , where $\omega_{\bf p}=\sqrt{{\bf p}^2+m^2}$ ; the choice of how the contour passes these poles selects the Green's function, with the Feynman prescription shifting $+\omega_{\bf p}$ below and $-\omega_{\bf p}$ above the real axis. Steering both poles the same way gives the retarded or advanced response of classical field theory. Feynman's choice splits them.

Figure

The

i\epsilon

prescription: how the poles are displaced and which one each time direction catches

The Stückelberg–Feynman picture interprets the negative-frequency contribution as antiparticle propagation forward in time, rather than literally as a particle travelling backward in time. The single propagator fuses the two separate diagrams of old hole theory into one analytic expression and dispenses entirely with the picture of an infinitely filled Dirac sea. But the leakage of $\Delta_F$ outside the light cone worries every careful student, and it should — until you see that no observable leaks. The causal content of a relativistic field theory is encoded in its commutators: the field commutator is the difference of the positive- and negative-frequency two-point functions and vanishes exactly at spacelike separation, $[\hat{\phi}(x),\hat{\phi}(y)]=0$ for $(x-y)^2<0$ . That cancellation is the second picture of this article, and it is the one worth burning into memory.

Figure

Microcausality: the commutator vanishes outside the light cone, though the propagator does not

The core equations

We start from the defining Green's-function equation, because everything downstream is a way of inverting it. The Feynman propagator is the specific inverse of the Klein-Gordon operator selected by the vacuum boundary condition. We work in the mostly-minus metric $(+,-,-,-)$ with $\hbar=c=1$ throughout, restoring units only in the numerical example.

(\Box_x + m^2)\,\Delta_F(x-y) = -\delta^4(x-y) \qquad (1)

where: $\Box=\partial_t^2-\nabla^2$ is the d'Alembertian; $m$ — the field quantum's rest mass; $\delta^4$ — the four-dimensional spacetime delta. Reading it physically: $\Delta_F$ is a Green's function, so it responds to a unit point source in spacetime; the homogeneous freedom (add any solution of $(\Box+m^2)\psi=0$ ) is precisely the pole-routing ambiguity that a boundary condition must resolve. The poles occur on the mass shell, $p^2=m^2$ , corresponding to $p^0=\pm\omega_{\bf p}$ . The ambiguity in defining the Green's function is resolved by the Feynman $+i\epsilon$ prescription, which specifies how the contour passes the poles.

\Delta_F(x-y)=\int\!\frac{\mathrm d^4 p}{(2\pi)^4}\,\frac{i\,e^{-ip\cdot(x-y)}}{p^2-m^2+i\epsilon} \qquad (2)

where: $p\cdot(x-y)=p^0(x^0-y^0)-\mathbf p\cdot(\mathbf x-\mathbf y)$ , $\epsilon\to0^+$ is infinitesimal and positive, and $\omega_{\mathbf p}=\sqrt{\mathbf p^2+m^2}$ . Reading it physically: the Feynman prescription shifts the positive-energy pole to $p^0=+\omega_{\mathbf p}-i\epsilon$ and the negative-energy pole to $p^0=-\omega_{\mathbf p}+i\epsilon$ . This pole displacement fixes the time-ordering prescription and, after the appropriate antiparticle reinterpretation for charged fields, gives the Stückelberg–Feynman picture. For the Dirac field the same denominator carries the numerator $\not p+m$ , but the pole structure and every causal statement below are identical. The time-ordered vacuum expectation value is what the diagram rule secretly computes, and it is the form that makes causality visible. Doing the $p^0$ integral first turns equation (2) into a sum of two on-shell pieces stitched together by time-ordering.

\Delta_F(x-y)=\theta(x^0-y^0)\,\Delta^+(x-y)+\theta(y^0-x^0)\,\Delta^-(x-y) \qquad (3)

where: $\theta$ is the Heaviside step; $\Delta^\pm(x-y)=\int\dfrac{\mathrm d^3p}{(2\pi)^3\,2\omega_p}\,e^{\mp i p\cdot(x-y)}\big|_{p^0=\omega_p}$ are the positive/negative-frequency Wightman functions. Reading it physically: for $x^0>y^0$ (later at $x$ ) only the positive-energy piece $\Delta^+$ contributes — a particle propagates forward; for $x^0<y^0$ the negative-frequency piece $\Delta^-$ contributes — an antiparticle propagates the other way. One object, both processes. The Lorentz-invariant measure $\mathrm d^3p/2\omega_p$ appearing here is the key to the causality proof.

Derivation AThe iε prescription splits time by residues

Our goal is equation (3): show that the momentum-space propagator (2), when the energy integral $\int\mathrm dp^0$ is done by contour, produces the time-ordered structure — positive-energy forward, negative-energy backward. The strategy is to write the denominator's two poles with their $+i\epsilon$ displacements, apply Jordan's lemma to decide which half-plane to close in each time direction, and read off the residues. Watch the sign of $x^0-y^0$ closely: it dictates the closure, and the closure dictates which pole (hence which physical process) survives.

Step 1. Factor the denominator near its poles. With $t\equiv x^0-y^0$ , $p^2-m^2+i\epsilon=(p^0)^2-\omega_p^2+i\epsilon=(p^0-\omega_p+i\epsilon')(p^0+\omega_p-i\epsilon')$ to first order, where $\epsilon'=\epsilon/2\omega_p>0$ . Why: expanding $(p^0)^2-(\omega_p-i\epsilon')^2=(p^0)^2-\omega_p^2+2i\omega_p\epsilon'$ reproduces $+i\epsilon$ with $\epsilon=2\omega_p\epsilon'$ ; so the pole at $+\omega_p$ sits just below the real axis and the pole at $-\omega_p$ just above it.

Step 2. For $t>0$ the factor $e^{-ip^0 t}$ decays in the lower half $p^0$ -plane, so close the contour downward (clockwise). The only enclosed pole is $p^0=+\omega_p-i\epsilon'$ . Why: Jordan's lemma requires closing where $e^{-ip^0 t}=e^{-i(\mathrm{Re}\,p^0)t}e^{+(\mathrm{Im}\,p^0)t}$ decays; for $t>0$ that needs $\mathrm{Im}\,p^0<0$ . Clockwise closure gives an extra $-2\pi i$ times the residue.

Step 3. Evaluate the residue at $p^0=\omega_p$ : the $\mathrm dp^0$ integral yields $\dfrac{i}{2\omega_p}e^{-i\omega_p t}$ (the $2\pi$ cancels against the measure and the $-2\pi i$ from clockwise closure combines with the explicit $i$ in (2)). Reinstating the $\mathrm d^3p$ integral gives $\theta(t)\,\Delta^+(x-y)$ . Why: the residue of $1/[(p^0-\omega_{\bf p})(p^0+\omega_{\bf p})]$ at $p^0=\omega_{\bf p}$ is $1/(2\omega_{\bf p})$ , so the surviving exponential $e^{-i\omega_{\bf p}t}$ is the positive-frequency, positive-energy contribution propagating forward in time. For a charged field this admits the particle interpretation.

Step 4. For $t<0$ , $e^{-ip^0 t}$ decays in the upper half-plane; close upward (counterclockwise), enclosing only $p^0=-\omega_p+i\epsilon'$ . Its residue gives $e^{+i\omega_p t}$ , i.e. $\theta(-t)\,\Delta^-(x-y)$ . Why: a wave $e^{+i\omega_p t}$ with $t<0$ is a negative-frequency solution running backward — the antiparticle. The two step-functions assemble exactly (3), so the $+i\epsilon$ is time-ordering. The displaced poles sit at $+\omega_p-i\epsilon'$ and $-\omega_p+i\epsilon'$ ; the residue at each is $1/2\omega_p$ ; and $t>0$ selects $e^{-i\omega_p t}$ while $t<0$ selects $e^{+i\omega_p t}$ .

Go deeper · Derivation B · 1 minMicrocausality: the commutator vanishes for spacelike separation

Derivation BMicrocausality: the commutator vanishes for spacelike separation

Here we prove the real content of causality: the field commutator $[\hat\phi(x),\hat\phi(y)]=\Delta(x-y)$ is a $c$ -number invariant function that is exactly zero whenever $(x-y)^2<0$ (spacelike). The strategy is to write $\Delta$ as the difference of the two Wightman functions, note that it is Lorentz invariant, and then use that invariance to rotate any spacelike separation into a pure-space configuration where the integrand is manifestly odd and integrates to zero. The one subtlety is that this rotation is only legal for spacelike intervals — the crux of the whole argument.

Step 1. Write the commutator as $\Delta(x-y)=\Delta^+(x-y)-\Delta^-(x-y)=\displaystyle\int\!\frac{\mathrm d^3p}{(2\pi)^3\,2\omega_p}\big(e^{-ip\cdot(x-y)}-e^{+ip\cdot(x-y)}\big)\Big|_{p^0=\omega_p}$ . Why: for a free field $[\hat\phi(x),\hat\phi(y)]$ reduces to a $c$ -number (the mode operators' $[a,a^\dagger]$ are numbers), and it is exactly the antisymmetric combination of positive- and negative-frequency parts.

Step 2. Note that the measure $\mathrm d^3p/2\omega_p$ and the phase $p\cdot(x-y)$ are each Lorentz invariant, so $\Delta(x-y)$ is a Lorentz scalar depending only on the invariant $(x-y)^2$ and on the sign of $x^0-y^0$ . Why: $\mathrm d^4p\,\delta(p^2-m^2)\theta(p^0)=\mathrm d^3p/2\omega_p$ is the standard invariant on-shell measure; invariance is what lets us change frames freely for spacelike arguments.

Step 3. For a spacelike interval choose the frame where $x^0-y^0=0$ and $(x-y)^2=-|\mathbf r|^2$ . Then $p\cdot(x-y)=-\mathbf p\cdot\mathbf r$ and $\Delta=\displaystyle\int\!\frac{\mathrm d^3p}{(2\pi)^3 2\omega_p}\big(e^{i\mathbf p\cdot\mathbf r}-e^{-i\mathbf p\cdot\mathbf r}\big)$ . Why: a spacelike separation always admits a frame with equal times (its invariant is negative); a timelike one never does — this is exactly where the argument would fail and correctly so.

Step 4. The integrand is odd under $\mathbf p\to-\mathbf p$ (the measure $\mathrm d^3p/2\omega_p$ is even, the bracket flips sign), so the integral vanishes: $\Delta(x-y)=0$ for $(x-y)^2<0$ . Why: an odd function integrated over all $\mathbf p$ gives zero. Hence $[\hat\phi(x),\hat\phi(y)]=0$ at spacelike separation — microcausality — even though $\Delta_F$ itself is nonzero there. Measurements commute, so none can signal to another outside the light cone. Concretely: $\Delta$ is the antisymmetric combination $\Delta^+-\Delta^-$ ; the equal-time spacelike integrand is odd in $\mathbf{p}$ and integrates to zero; a regulated 1D analogue integrates to $\approx 0$ for spacelike separation and to a nonzero value for timelike separation.

Limiting cases

case	result	meaning
$x^0>y^0$	$\Delta_F\to\Delta^+$ , particle forward	only the $+\omega_p$ residue survives
$x^0<y^0$	$\Delta_F\to\Delta^-$ , negative-frequency contribution	only the $-\omega_p$ residue survives
spacelike $(x-y)^2<0$	$\Delta_F\ne0$ but $\Delta=[\phi,\phi]=0$	no signaling; microcausality holds
$c\to\infty$ (nonrelativistic)	recovers $K$ of The Propagator from Path Integrals	mass shell $E=p^2/2m$ , single pole
$\epsilon\to0^+$ with sign flipped	retarded/advanced Green's function	classical response, not the vacuum choice

Common trapCausality does NOT mean the propagator vanishes outside the light cone

The single most persistent error is to expect $\Delta_F(x-y)=0$ for spacelike separation and to panic when it is not. The Feynman propagator is generally nonzero for spacelike separations, with its magnitude falling exponentially at large spacelike distance. This spacelike tail does not represent physical particles, nor does it permit signals to propagate outside the light cone. This is not an acausality, because $\Delta_F$ is not an observable. The observable content lives in the commutator $[\hat\phi(x),\hat\phi(y)]=\Delta(x-y)$ , and it is the commutator that vanishes for $(x-y)^2<0$ . Confusing "propagator support" with "causality" leads to imaginary paradoxes; the correct statement — microcausality — is that spacelike-separated measurements commute, and that is what forbids superluminal signaling.

Worked examples

Example 1 — the residue that makes a particle go forward. Show explicitly that for $t=x^0-y^0>0$ the $p^0$ integral in equation (2) produces the positive-energy amplitude $\dfrac{1}{2\omega_p}e^{-i\omega_p t}$ .

Step 1. Isolate the energy integral: $I=\displaystyle\int_{-\infty}^{\infty}\frac{\mathrm dp^0}{2\pi}\,\frac{i\,e^{-ip^0 t}}{(p^0-\omega_p+i\epsilon')(p^0+\omega_p-i\epsilon')}$ with $\epsilon'>0$ .

Step 2. For $t>0$ close downward; the enclosed pole is $p^0=\omega_p-i\epsilon'$ . Clockwise closure contributes $-2\pi i\times(\text{residue})$ .

Step 3. The residue is $\dfrac{i\,e^{-i\omega_p t}}{2\omega_p}$ (evaluating the second factor at $p^0=\omega_p$ : $2\omega_p$ ). So $I=\dfrac{1}{2\pi}\cdot(-2\pi i)\cdot\dfrac{i\,e^{-i\omega_p t}}{2\omega_p}=\dfrac{-2\pi i\cdot i}{2\pi}\cdot\dfrac{e^{-i\omega_p t}}{2\omega_p}$ .

Step 4. Simplify $-2\pi i\cdot i=-2\pi i^2=+2\pi$ , so $I=\dfrac{2\pi}{2\pi}\cdot\dfrac{e^{-i\omega_p t}}{2\omega_p}=\mathbf{\dfrac{e^{-i\omega_p t}}{2\omega_p}}$ . A pure positive-energy wave, forward in time — precisely $\Delta^+$ before the $\mathrm d^3p$ integration.

Example 2 — the spacelike commutator cancels; the timelike one does not. Using a tractable regulated 1D toy version $\Delta(r,t)=\displaystyle\int_{-\infty}^{\infty}\frac{\mathrm dp}{2\pi\,2\omega_p}\big(e^{i(pr-\omega_p t)}-e^{-i(pr-\omega_p t)}\big)$ with $\omega_p=\sqrt{p^2+m^2}$ , evaluate the equal-time ( $t=0$ ) case and contrast a timelike ( $r=0$ ) case, taking $m=1$ .

Step 1. Equal time $t=0$ : the bracket is $e^{ipr}-e^{-ipr}=2i\sin(pr)$ , and $\Delta(r,0)=\displaystyle\int\frac{\mathrm dp}{2\pi\,2\omega_p}\,2i\sin(pr)$ .

Step 2. The integrand $\dfrac{\sin(pr)}{2\omega_p}$ is odd in $p$ (numerator odd, $\omega_p$ even), so the symmetric integral gives $\Delta(r,0)=\mathbf{0}$ — the spacelike (here purely spatial) commutator vanishes.

Step 3. Timelike $r=0$ : the bracket is $e^{-i\omega_p t}-e^{+i\omega_p t}=-2i\sin(\omega_p t)$ , and $\Delta(0,t)=-\displaystyle\int\frac{\mathrm dp}{2\pi\,2\omega_p}\,2i\sin(\omega_p t)$ .

Step 4. Now the integrand $\dfrac{\sin(\omega_p t)}{2\omega_p}$ is even in $p$ (both $\omega_p$ and $\sin(\omega_p t)$ even), so the integral does not vanish. Numerically at $t=1$ , $m=1$ a regulated evaluation gives $\Delta(0,1)\approx\mathbf{-0.4i}$ (nonzero). The commutator lives inside the cone and dies outside it — microcausality, made arithmetic.

Origins · 1 minFeynman propagator and how this was discovered

Feynman propagator

1941–1949 · The causal Green's function of relativistic QFT, developed by Stückelberg and Feynman: particles forward, antiparticles backward in time

The propagator's causal structure was first written down by Ernst Stückelberg in 1941 (Helv. Phys. Acta 14, 588), who proposed reading negative-energy solutions as antiparticles moving backward in time and organized the amplitudes accordingly. Richard Feynman independently rediscovered and popularized the same object in 1949 (Phys. Rev. 76, 769), where it became the internal-line rule of his diagrams. The $+i\epsilon$ prescription is the analytic partner of Wick's 1954 rotation to imaginary time; the vanishing of the field commutator at spacelike separation traces to the invariant $\Delta$ -function written by Jordan and Pauli in 1928. Together these fix the propagator as a Lorentz-invariant time-ordered Green's function of the wave operator.

History

1928

— Jordan and Pauli construct the invariant commutator function Δ(x), vanishing outside the light cone

1934

— Pauli and Weisskopf quantize the scalar field, reinterpreting its indefinite density as charge

1941

— Stückelberg proposes antiparticles as negative-energy solutions moving backward in time

1948

— Wheeler and Feynman's absorber theory sharpens the forward/backward propagation idea

1949

— Feynman publishes the propagator as the internal-line rule of his spacetime diagrams

1954

— Wick's rotation t → −iτ makes the iε prescription an analytic continuation to Euclidean time

Going further

This article builds the causal Green's function on top of the nonrelativistic kernel and composition law of The Propagator from Path Integrals, uses the $+i\epsilon$ as the analytic-continuation partner of the Wick rotation from Imaginary-Time and Euclidean Path Integrals, and places its pole on the mass shell of The Klein-Gordon Equation. The microcausality proved here — a spacelike commutator that vanishes only when the field is quantized with commutators for the scalar — is exactly the ingredient the track's closing article, Spin-Statistics and the CPT Theorem, turns into a theorem: choose the wrong statistics and this cancellation fails. Downstream, the propagator is the internal-line rule from which the LSZ reduction formula and the full interacting S-matrix are assembled — the machinery of quantum electrodynamics that lies just beyond this track. Note on conventions: the momentum-space propagator is $\tilde\Delta_F=i/(p^2-m^2+i\epsilon)$ — the explicit factor of $i$ in the numerator is required by the time-ordered VEV convention $\Delta_F=\langle 0|T\hat\phi\hat\phi|0\rangle$ , and the Green's-function normalization is $(\Box+m^2)\Delta_F=-\delta^4$ , as in equation (1) and equation (2) above.

Prerequisites. The Klein-Gordon Equation · The Propagator from Path Integrals · Imaginary-Time and Euclidean Path Integrals · Negative-Energy Solutions and the Interpretive Crisis. Leads to. Spin-Statistics and the CPT Theorem · the LSZ reduction formula · the interacting S-matrix · quantum electrodynamics.

Why is there a factor of i in the numerator of the momentum-space propagator?

It comes from defining the propagator as the time-ordered vacuum expectation value $\Delta_F=\langle 0|T\hat\phi(x)\hat\phi(y)|0\rangle$ . Doing the contour integral of $1/(p^2-m^2+i\epsilon)$ produces a factor of $2\pi i$ from the residue, and matching to the on-shell Wightman functions $\Delta^\pm$ (which carry no explicit $i$ ) requires the momentum-space form to be $i/(p^2-m^2+i\epsilon)$ . Some texts absorb the $i$ into the definition of the Green's function of equation (1) instead; the physics — the pole positions and the causal structure — is identical either way, but the diagram Feynman rules conventionally put the $i$ in the propagator.

What is the difference between the Feynman, retarded, and advanced propagators?

All three invert the same operator $(\Box+m^2)$ and differ only in how the contour skirts the two poles at $\pm\omega_p$ . The retarded propagator pushes both poles below the real axis, so it vanishes for $t<0$ (pure future response — the classical Green's function). The advanced one pushes both above, vanishing for $t>0$ . The Feynman propagator splits them: $+\omega_p$ below, $-\omega_p$ above. That split is the only choice consistent with the vacuum being the lowest-energy state, and it is what makes $\Delta_F$ Lorentz invariant while the retarded/advanced functions single out a time direction.

If Δꜰ is nonzero outside the light cone, doesn't that allow faster-than-light influence?

No, because $\Delta_F$ is not measurable on its own. Any physical influence of a measurement at $y$ on one at $x$ is governed by the commutator $[\hat\phi(x),\hat\phi(y)]$ , and that commutator is exactly zero for spacelike separation, as the second derivation shows. The nonzero spacelike value of $\Delta_F$ describes virtual propagation that always cancels out of any observable. Concretely, the amplitude for a real signal factorizes through commutators of local operators, so the spacelike leakage of the propagator never appears in a cross-section or an expectation value. Microcausality, not propagator support, is the operative principle.

How does the Feynman propagator reduce to the nonrelativistic path-integral kernel?

Take the nonrelativistic limit by writing $\omega_p=\sqrt{\mathbf p^2+m^2}\approx m+\mathbf p^2/2m$ and factoring out the rest-energy phase $e^{-imt}$ . The negative-energy pole decouples (it costs energy $\sim 2m$ to excite), leaving a single pole and a Green's function that propagates only forward in time. What remains is exactly the Schrödinger kernel $K$ with the free dispersion $E=\mathbf p^2/2m$ — the object whose composition law The Propagator from Path Integrals derives. The relativistic propagator is thus the parent object; the nonrelativistic kernel is its low-energy, single-pole child.

Why does the causality argument work only for spacelike, not timelike, separations?

Because the frame change at the heart of the proof is legal only for spacelike intervals. A spacelike separation has $(x-y)^2<0$ and can always be boosted to a frame with $x^0=y^0$ (equal times), where the mode integrand is odd in $\mathbf p$ and integrates to zero. A timelike separation has $(x-y)^2>0$ ; no boost can equalize the times, and in fact the ordering of $x^0$ and $y^0$ is Lorentz-invariant there. So the commutator should be nonzero inside the cone — that is where physical propagation happens — and the proof correctly refuses to make it vanish. The argument is not a trick that "accidentally" works; it precisely tracks the causal structure of spacetime.

What does microcausality have to do with spin and statistics?

Everything, and it is the reason this article precedes the spin-statistics theorem. The spacelike commutator vanishes for the scalar field only because we quantized it with commutators $[a,a^\dagger]$ ; the two mode integrals then subtract and cancel. If you instead quantized the scalar with anticommutators, the commutator function would become an anticommutator combination that does not cancel outside the cone — microcausality would fail. For a spin- $\tfrac12$ field the arithmetic is reversed: commutators break causality, anticommutators save it. So demanding a vanishing spacelike (anti)commutator forces bosons to commute and fermions to anticommute — the spin-statistics connection, developed in the next article.

Is the iε prescription related to the Wick rotation?

Yes — they are two faces of the same analytic continuation. The Wick rotation $t\to-i\tau$ of Imaginary-Time and Euclidean Path Integrals rotates the time contour by ninety degrees; the $+i\epsilon$ is the infinitesimal tilt of the energy contour that makes that rotation well-defined and single-valued. Rotating $t$ to $-i\tau$ turns the oscillating Minkowski denominator $p^2-m^2+i\epsilon$ into the Euclidean $p_E^2+m^2$ , which has no poles on the real axis at all — the reason Euclidean propagators are unambiguous. The $i\epsilon$ tells you which way to rotate back, so the Feynman propagator is the boundary value of the Euclidean one approached from the correct side.