13.09.02 · gr-cosmology / microlocal-qft-curved-spacetimes

Klein-Gordon equation on a globally hyperbolic spacetime

shipped3 tiersLean: none

Anchor (Master): Gérard, Microlocal Analysis of Quantum Fields on Curved Spacetimes (EMS, 2019), Ch. 5-6; Bär-Ginoux-Pfäffle (EMS, 2007), Ch. 3; Wald, Quantum Field Theory in Curved Spacetime and Black Hole Thermodynamics (Chicago, 1994), Ch. 3; Hörmander, The Analysis of Linear Partial Differential Operators III (Springer, 1985), Ch. XXIII

Intuition Beginner

The Klein-Gordon equation is the simplest relativistic wave equation. In flat spacetime it says that the d'Alembertian of a scalar field $ϕ$ plus the field times its squared mass $m^{2}$ is zero. The d'Alembertian is the time-and-space version of the Laplacian — a second-derivative operator that subtracts the time-second-derivative from the spatial-second-derivative. The equation describes a free scalar field, the simplest relativistic version of "a single quantity that fills space and time and obeys a wave-like rule." Photons, gravitons, and Standard Model bosons have more structure; the Klein-Gordon field is the bare prototype that the higher-spin generalisations decorate.

On a curved Lorentzian spacetime $(M, g)$ , the same equation makes sense once you replace flat-spacetime derivatives by their covariant analogues. The d'Alembertian gets a curved-spacetime version $□_{g}$ built from the metric and its connection, and the equation reads $(□_{g} + m^{2}) u = 0$ . The mass parameter is unchanged; the geometry enters through $□_{g}$ . On Minkowski with the flat metric this reduces to the flat-space equation. On Schwarzschild or de Sitter or FLRW it produces a new equation whose solutions encode wave propagation through whatever curved geometry you picked.

The strength of the Klein-Gordon equation on a globally hyperbolic spacetime is that it has a clean initial-value problem. You specify a Cauchy hypersurface $Σ$ — a "moment-of-time slice" guaranteed to exist by the Geroch-Bernal-Sánchez theorem from 13.09.01 — and you give two pieces of initial data on it: the value of the field and the time-derivative of the field, both smooth and supported in a bounded region. There is then exactly one smooth solution of the equation on the whole spacetime that matches this data on $Σ$ . The support of the solution stays inside the union of the future and past light cones of the support of the data — disturbances propagate at the speed of light or slower.

This is the cleanest existence-and-uniqueness statement in classical relativistic field theory. It is the relativistic analogue of the Newtonian story for a vibrating string: specify the initial shape and the initial velocity at every point of the string, and the string's motion is determined for all later time. The mathematical machinery is more elaborate — energy estimates over Cauchy slices, a global patching argument due to Leray and Choquet-Bruhat, propagation-of-singularities theorems for the wave operator — but the human content is the same.

Two pieces of structure come along for the ride. Retarded and advanced Green's operators $E^{+}$ and $E^{-}$ act on a source $f$ supported in a compact region. The retarded operator returns the unique solution supported in the future of the source; the advanced operator returns the unique solution supported in the past of the source. Their difference $E = E^{-} - E^{+}$ is the causal propagator or Pauli-Jordan commutator, named for the 1928 paper by Jordan and Pauli that introduced it in flat-space relativistic electromagnetism. The causal propagator encodes the entire symplectic structure on the phase space of Cauchy data and is the basic ingredient of canonical quantisation of the free scalar field.

Without global hyperbolicity, none of this works. The wave equation on a spacetime with closed timelike curves (such as Gödel) would have to satisfy a periodicity condition along the loops, and generic initial data is inconsistent. The wave equation on anti-de Sitter does not have unique solutions from initial data alone — you must also specify boundary conditions at the conformal infinity that null geodesics reach in finite affine time. The geometric condition from 13.09.01 is exactly what makes the wave equation behave: a Cauchy hypersurface plus finite propagation speed plus a smooth metric splitting in which the d'Alembertian becomes a manifestly strictly hyperbolic operator.

Visual Beginner

The picture to hold is a Cauchy hypersurface with initial data and the resulting solution constrained to lie inside the causal future and past of the data's support.

The Cauchy data $(u_{0}, u_{1})$ live on $Σ$ . The retarded fundamental solution $E^{+}$ is the integral operator that takes a source $f$ supported in a region and returns the unique solution of $(□_{g} + m^{2}) u = f$ supported in the causal future of that region. The advanced operator $E^{-}$ does the same in the past direction. Their difference $E = E^{-} - E^{+}$ is supported in the union of the future and past, with the symmetry $E (x, y) = - E (y, x)$ — the antisymmetry that makes $E$ the integral kernel of a symplectic form on Cauchy data.

The picture you should not draw, but should keep in mind as a counterpoint, is the same setup on an anti-de Sitter background. Null geodesics reach the timelike conformal boundary in finite affine parameter; data flowing in from the boundary cannot be predicted from interior initial data alone. The Cauchy problem fails for the cleanest possible reason: there is no Cauchy hypersurface.

Worked example Beginner

Take Minkowski spacetime with coordinates $(t, x, y, z)$ , light-speed $c = 1$ , and the standard flat metric. The d'Alembertian is the operator that takes a smooth field and returns the spatial Laplacian minus the second time derivative. Consider the free massless Klein-Gordon equation (mass parameter $m = 0$ ) with initial data on the Cauchy slice at $t = 0$ : a smooth bump function $u_{0}$ for the value of the field, and a smooth bump function $u_{1}$ for the time derivative of the field, both supported inside a spatial ball of radius $R$ centred on the origin.

Step 1. Decompose into spatial frequencies. Take the Fourier transform of the field in the three spatial coordinates while leaving $t$ alone. The wave equation becomes, for each spatial frequency $k$ , a single ordinary differential equation in time: the time-second-derivative of the Fourier-transformed field equals the squared spatial-frequency magnitude $∣ k ∣^{2}$ times the negative of the field. This is the harmonic-oscillator equation with angular frequency $∣ k ∣$ .

Step 2. Solve each frequency mode. The general solution is a linear combination of $cos (∣ k ∣ t)$ and $sin (∣ k ∣ t) /∣ k ∣$ . The two initial-data Fourier transforms pin down the two coefficients uniquely: the cosine coefficient is the Fourier transform of $u_{0}$ , and the sine coefficient is the Fourier transform of $u_{1}$ .

Step 3. Reconstruct the solution. Invert the Fourier transform on the time-evolved modes. This yields a closed-form solution: $u (t, x)$ as a spatial convolution of $u_{0}$ and $u_{1}$ against the cosine and sine kernels respectively.

Step 4. Verify finite propagation speed. The cosine-kernel and the sine-kernel are both spatially supported on a sphere of radius $t$ around the origin (a standard spherical-mean computation due to Kirchhoff). Convolving against $u_{0}$ and $u_{1}$ supported in a ball of radius $R$ , the solution $u (t, x)$ vanishes for $∣ x ∣$ outside a ball of radius $R + ∣ t ∣$ . The disturbance has spread at the speed of light, no faster, exactly as the finite-propagation-speed property predicts.

Step 5. The retarded propagator $E^{+}$ is the unique distribution on Minkowski supported in the closed future light cone of the origin that satisfies the wave equation with a unit point source at the origin. In the massless case, $E^{+}$ is supported exactly on the future light cone boundary ${t = ∣ x ∣}$ as a delta-distribution against the Minkowski squared distance, multiplied by the Heaviside step in time. The advanced propagator $E^{-}$ is the time-reverse, supported on the past light cone boundary. Their difference $E = E^{-} - E^{+}$ is supported on the full light cone (past and future) and is the Pauli-Jordan commutator.

Step 6. Massive case. With $m > 0$ the same Fourier-transform procedure works, replacing $∣ k ∣$ by $ω_{k} = ∣ k ∣^{2} + m^{2}$ . The retarded propagator gains an extra term inside the future light cone — a Bessel-function tail proportional to $J_{1} (m x^{μ} x_{μ})$ — on top of the delta-on-the-cone piece. The new term is the wake of the massive wave: a massive Klein-Gordon disturbance, unlike a massless one, leaves a residue in its interior wake that decays at large times.

What this tells us: the Klein-Gordon equation on Minkowski is solvable in closed form by Fourier transform, and the explicit propagator $E^{+}$ has its support exactly in the closed future light cone. This is the simplest case of the general theorem on a globally hyperbolic background; the curved-spacetime story replaces the explicit Fourier calculation with an energy-estimate-plus-patching argument due to Leray, Choquet-Bruhat, and (in the modern textbook formulation) Bär-Ginoux-Pfäffle, but produces propagators with the same support structure and antisymmetric difference $E = E^{-} - E^{+}$ .

Check your understanding Beginner

Exercise (easy, multiple choice).

On a globally hyperbolic spacetime $(M, g)$ , the Cauchy problem for $(□_{g} + m^{2}) u = 0$ specifies which data on a Cauchy hypersurface $Σ$ ?

A. The value of $u$ on $Σ$ only B. The value of $u$ and its time derivative on $Σ$ C. The value of $u$ at all points of $M$ D. The value of $u$ at one point of $Σ$ and its full Taylor series there

Hint

The Klein-Gordon equation is second-order in time. Compare with the harmonic-oscillator equation $\overset{q}{¨} = - ω^{2} q$ — how many initial conditions does that need?

Answer

B. The value of $u$ and its time derivative on $Σ$ .

The Klein-Gordon equation is second-order in time, so its initial-value problem requires both the field value on the slice and the time derivative of the field along the future-directed unit normal to the slice. With both data the solution is unique throughout $M$ . Specifying only the field value (option A) leaves the time derivative free and produces a one-parameter family of solutions. Specifying the field everywhere on $M$ (option C) is over-determined. Option D is a Cauchy-Kowalevski style local condition that does not generalise to a global statement on a Lorentzian manifold.

Exercise (easy, true/false).

A smooth solution $u$ of $(□_{g} + m^{2}) u = 0$ on a globally hyperbolic spacetime with compactly supported initial data on a Cauchy hypersurface $Σ$ must vanish outside the causal future-and-past of the support of the data.

Hint

This is the finite-propagation-speed property of the Klein-Gordon equation. Recall that disturbances travel at the speed of light or slower.

Answer

True.

The classical Cauchy theorem on a globally hyperbolic background says the support of $u$ lies inside the causal future-and-past of the support of the data, written $J (supp (u_{0}) \cup supp (u_{1}))$ . The mechanism is the energy-estimate argument: multiplying the equation by the time derivative of $u$ , integrating over a slab between two Cauchy slices and using the Gauss theorem, gives a bound on the energy norm over a later slice in terms of the energy norm on the earlier slice. If the initial energy is supported in a compact region $K$ , the propagated energy at time $t$ is supported in the intersection of $J (K)$ with the later slice.

Exercise (medium, multiple choice).

The Pauli-Jordan commutator $E = E^{-} - E^{+}$ on a globally hyperbolic spacetime is

A. A smooth function on $M \times M$ B. A distribution on $M \times M$ supported on pairs $(x, y)$ with $y \in J (x)$ C. The Feynman propagator D. Identically zero

Hint

The retarded propagator $E^{+}$ is supported in pairs with $y \in J^{+} (x)$ ; the advanced $E^{-}$ in pairs with $y \in J^{-} (x)$ . Take the difference.

Answer

B. A distribution on $M \times M$ supported on pairs $(x, y)$ with $y \in J (x)$ .

The retarded propagator $E^{+}$ is supported where the second argument lies in the causal future of the first; the advanced $E^{-}$ is supported where the second lies in the causal past. Their difference $E = E^{-} - E^{+}$ has support in the causal future-and-past of the diagonal. On Minkowski, $E$ is non-zero on the full light cone, vanishes inside the cone for the massless equation, and has Bessel-function support inside the cone for the massive equation.

Option A is wrong because $E$ is a distribution, not a function — the propagators have delta-type singularities on the light cone. Option C is wrong because the Feynman propagator is a particular complex linear combination of $E^{+}$ and $E^{-}$ selected by an $i ϵ$ prescription, not equal to $E$ . Option D is the degenerate case $m = 0$ in one spacetime dimension; in physical dimensions $E$ is non-zero.

Exercise (medium, numeric).

Consider the massless Klein-Gordon equation on Minkowski with initial data: the field is zero everywhere on the slice $t = 0$ , and the time derivative on that slice is a bump function supported inside the spatial ball $∣ x ∣ \leq 2$ . At time $t = 3$ , what is the smallest spatial radius $R^{*}$ such that the field $u (3, x)$ is zero for all $∣ x ∣ > R^{*}$ ? Give the numerical value.

Hint

Finite-propagation-speed in Minkowski: a disturbance starting in $∣ x ∣ \leq R_{0}$ at $t = 0$ reaches at most $∣ x ∣ \leq R_{0} + ∣ t ∣$ at time $t$ . Apply with $R_{0} = 2$ and $t = 3$ .

Answer

$R^ = 5$.*

By finite propagation speed for the d'Alembertian on Minkowski, the support of $u (3, \cdot)$ lies in the spatial ball of radius $R_{0} + ∣ t ∣ = 2 + 3 = 5$ around the origin. Outside that ball, the field is zero. The value $R^{*} = 5$ is sharp: for generic non-zero initial-velocity data supported on the full ball $∣ x ∣ \leq 2$ , the solution at $t = 3$ has support reaching exactly out to $∣ x ∣ = 5$ .

Formal definition Intermediate+

Throughout this unit $(M, g)$ denotes a smooth four-dimensional connected time-oriented globally hyperbolic Lorentzian manifold in the mostly-plus signature $(-, +, +, +)$ . By the Bernal-Sánchez theorem 13.09.01 we may identify $M$ with $R \times Σ$ via a smooth diffeomorphism in which the metric takes the form $g = - β^{2} d t^{2} + h_{t}$ , with $β : M \to R_{> 0}$ a smooth lapse and $h_{t}$ a smooth $t$ -varying Riemannian metric on $Σ$ . The signature convention reverses signs throughout if one prefers $(+, -, -, -)$ ; the geometric content is identical. Adopting the convention of Gérard 2019 and Bär-Ginoux-Pfäffle 2007 simplifies the comparison with the modern Lorentzian-PDE literature.

The d'Alembertian (or wave operator, or scalar Laplacian) on $(M, g)$ is the second-order linear differential operator

□_{g} := g^{μν} \nabla_{μ} \nabla_{ν} = \frac{1}{∣ g ∣} \partial_{μ} (∣ g ∣ g^{μν} \partial_{ν}),

where $\nabla$ is the Levi-Civita connection of $g$ , $∣ g ∣ = ∣ det g_{μν} ∣$ , and the second formula expresses $□_{g}$ in any chart. In mostly-plus signature on Minkowski, $□ = - \partial_{t}^{2} + \partial_{x}^{2} + \partial_{y}^{2} + \partial_{z}^{2}$ . In the unit-lapse Bernal-Sánchez gauge $g = - d t^{2} + h_{t}$ , the d'Alembertian reads $□_{g} u = - \partial_{t}^{2} u - (\partial_{t} ln det h_{t}) \partial_{t} u + Δ_{h_{t}} u$ , where $Δ_{h_{t}}$ is the spatial Laplace-Beltrami operator on $(Σ, h_{t})$ . The negative sign in front of $\partial_{t}^{2}$ is the hallmark of mostly-plus signature: $□_{g}$ is normally hyperbolic with principal symbol $σ_{2} (□_{g}) (x, ξ) = g^{μν} ξ_{μ} ξ_{ν}$ , equal to the squared Lorentzian length of the covector $ξ$ .

The Klein-Gordon equation for a real (or complex) scalar field $u : M \to R$ with mass $m \geq 0$ and an optional smooth potential $V : M \to R$ is

P u := (□_{g} + m^{2} + V (x)) u = f, u \in C^{\infty} (M),

where $f \in C_{c}^{\infty} (M)$ is a smooth source of compact support. The homogeneous equation is the case $f = 0$ . The notation $P = □_{g} + m^{2} + V$ for the Klein-Gordon operator is standard. The principal symbol of $P$ agrees with that of $□_{g}$ — the lower-order $m^{2} + V$ terms do not contribute — so $P$ is again normally hyperbolic.

Given a smooth spacelike Cauchy hypersurface $Σ \subset M$ with future-directed unit normal vector field $n$ , the Cauchy data of a smooth solution $u$ on $Σ$ is the pair

(u_{0}, u_{1}) := (u ∣_{Σ}, n^{μ} \nabla_{μ} u ∣_{Σ}) \in C^{\infty} (Σ) \oplus C^{\infty} (Σ) .

The Cauchy problem for the Klein-Gordon equation is the problem: given $(u_{0}, u_{1}) \in C_{c}^{\infty} (Σ) \oplus C_{c}^{\infty} (Σ)$ and a source $f \in C_{c}^{\infty} (M)$ , find $u \in C^{\infty} (M)$ with $P u = f$ , $u ∣_{Σ} = u_{0}$ , and $n^{μ} \nabla_{μ} u ∣_{Σ} = u_{1}$ .

A retarded fundamental solution of $P$ is a continuous linear map $E^{+} : C_{c}^{\infty} (M) \to C^{\infty} (M)$ satisfying

P E^{+} f = f, E^{+} P f = f on C_{c}^{\infty} (M), supp (E^{+} f) \subset J^{+} (supp f) .

The advanced fundamental solution $E^{-}$ is defined symmetrically with $J^{-}$ in place of $J^{+}$ . The causal propagator (or Pauli-Jordan commutator) is

E := E^{-} - E^{+} : C_{c}^{\infty} (M) \to C^{\infty} (M),

with support contained in $J (supp f) = J^{+} (supp f) \cup J^{-} (supp f)$ . The minus sign is the convention of Gérard 2019; many physics-side sources (e.g. Bär-Ginoux-Pfäffle 2007 §3.4 in one place, Wald 1994 §3.3) use the opposite sign $E = E^{+} - E^{-}$ . The two conventions agree up to an overall sign that propagates into the CCR commutator $[\hat{ϕ} (f), \hat{ϕ} (g)] = i ℏ E (f, g) 1$ ; choose one convention per text and stick with it.

The bilinear form $(u, v) \mapsto E (u, v) := ⟨ u, E v ⟩_{M}$ , integrated against the natural volume form $d vol_{g}$ , is antisymmetric: $E (u, v) = - E (v, u)$ . This is the symplectic form on the space of Cauchy data that powers canonical quantisation; see the Connections section.

Counterexamples to common slips

The Klein-Gordon equation $(□_{g} + m^{2}) u = 0$ on a spacetime that is not globally hyperbolic generically has no well-posed Cauchy problem. Anti-de Sitter is the canonical failure: solutions are not determined by interior initial data because null geodesics escape to the timelike conformal boundary in finite affine time and information leaks back in. Curing the failure requires imposing boundary conditions at conformal infinity by hand — the AdS/CFT dictionary supplies them in the holographic context.
The retarded propagator $E^{+}$ is not the same as the Feynman propagator $G_{F}$ even at the level of supports. $E^{+}$ has support in ${y \in J^{+} (x)}$ — a real light cone in the future. The Feynman propagator $G_{F} = i (E^{+} + E^{-}) /2 + positive-frequency boundary$ has its singular support on the full light cone but additionally carries a positive-frequency boundary prescription (an $i ϵ$ in flat space, a Hadamard state on a curved background) that selects which combination of modes propagates positively.
"Compactly supported initial data" is necessary for the simplest statement of the existence theorem. The general Cauchy problem with non-compact data is still well-posed on a globally hyperbolic background, but one must keep track of spatial decay rates (Sobolev spaces $H_{loc}^{s} (Σ)$ or $H^{s} (Σ_{t})$ with appropriate weights); the support property $supp (u) \subset J (supp (data))$ then needs reformulation as a domain-of-dependence statement.
The minus-sign convention $E = E^{-} - E^{+}$ versus $E = E^{+} - E^{-}$ is a perennial source of confusion. The unit adopts Gérard 2019's convention $E = E^{-} - E^{+}$ , which has the antisymmetric kernel $E (x, y) = - E (y, x)$ . The opposite convention $E^{+} - E^{-}$ also has antisymmetric kernel and merely flips an overall sign in the CCR commutator.
A solution $u$ of $(□_{g} + m^{2}) u = 0$ with $u ∣_{Σ} = 0$ and $\partial_{t} u ∣_{Σ} = 0$ on a Cauchy hypersurface is identically zero on all of $M$ . This is the uniqueness statement and follows from the energy estimate plus the global hyperbolicity of $(M, g)$ . Without global hyperbolicity, the uniqueness can fail.

Key derivation Intermediate+

Theorem (Cauchy problem for the Klein-Gordon equation; Leray 1953 / Choquet-Bruhat 1952 / Bär-Ginoux-Pfäffle 2007). Let $(M, g)$ be a connected time-oriented globally hyperbolic Lorentzian manifold and let $Σ \subset M$ be a smooth spacelike Cauchy hypersurface with future-directed unit normal $n$ . Let $P = □_{g} + m^{2} + V$ with $V \in C^{\infty} (M)$ and $m \geq 0$ . For every $(u_{0}, u_{1}, f) \in C_{c}^{\infty} (Σ) \oplus C_{c}^{\infty} (Σ) \oplus C_{c}^{\infty} (M)$ there exists a unique $u \in C^{\infty} (M)$ with

P u = f, u ∣_{Σ} = u_{0}, n^{μ} \nabla_{μ} u ∣_{Σ} = u_{1},

and

supp (u) \subset J (supp (u_{0}) \cup supp (u_{1}) \cup supp (f)) .

The solution $u$ depends continuously on $(u_{0}, u_{1}, f)$ in the natural $C^{\infty}$ topologies. As a consequence, there exist unique retarded and advanced fundamental solutions $E^{\pm} : C_{c}^{\infty} (M) \to C^{\infty} (M)$ for $P$ , with $supp (E^{\pm} f) \subset J^{\pm} (supp f)$ .

This is the foundational existence-and-uniqueness statement for the Klein-Gordon equation in the curved-spacetime setting. The proof rests on three ingredients: an energy estimate over a Cauchy slice, uniqueness via the energy estimate, and a global patching argument that turns local solutions into a global one. The canonical textbook statement appears in Bär-Ginoux-Pfäffle 2007 ^{[Bär-Ginoux-Pfäffle 2007 §3.1-3.4]} and Gérard 2019 ^{[Gérard 2019 §5.3-5.4]}.

The 1953 mimeographed Leray notes ^{[Leray 1953]} and the 1952 Choquet-Bruhat paper ^{[Choquet-Bruhat 1952]} established the local case in their own conventions. The global statement on a Lorentzian background became standard after Hawking-Ellis 1973 ^{[Hawking-Ellis 1973 §7.4]} and Friedlander 1975 ^{[Friedlander 1975]} consolidated the picture, with the algebraic-QFT statement of $E^{\pm}$ as Green's operators in Dimock 1980 ^{[Dimock 1980]} and Wald 1994 ^{[Wald 1994 §3.2]}.

Proof sketch (existence + uniqueness via energy estimate + global patching).

Step 1: Energy estimate. Work in the Bernal-Sánchez splitting $M ≅ R \times Σ$ with $g = - β^{2} d t^{2} + h_{t}$ . For a smooth function $u : M \to R$ , define the energy across the slice $Σ_{t} := {t} \times Σ$ by

E_{t} [u] := \int_{Σ_{t}} (∣ \partial_{t} u ∣^{2} + h_{t}^{ij} \partial_{i} u \partial_{j} u + (m^{2} + ∣ V ∣) u^{2}) β d vol_{h_{t}} .

A direct computation using $P u = f$ and the divergence theorem on the slab $[t_{1}, t_{2}] \times Σ$ yields, after a Grönwall application against a constant depending on $sup_{t \in [t_{1}, t_{2}]} ∥\nabla V ∥_{C^{0} (Σ_{t})}$ and on the lapse $β$ ,

E_{t_{2}} [u] \leq C (E_{t_{1}} [u] + \int_{t_{1}}^{t_{2}} ∥ f (s) ∥_{L^{2} (Σ_{s})}^{2} d s) .

The energy estimate is the analytic engine; the constant $C$ depends on the geometry and the slab but not on $u$ .

Step 2: Uniqueness. Suppose $u_{1}, u_{2}$ are two smooth solutions of the same Cauchy problem $(P u = f, u ∣_{Σ} = u_{0}, n^{μ} \nabla_{μ} u ∣_{Σ} = u_{1})$ . Their difference $w := u_{1} - u_{2}$ satisfies the homogeneous equation $P w = 0$ with zero Cauchy data on $Σ$ . Apply the energy estimate to $w$ on the slab $[0, t]$ : $E_{t} [w] \leq C E_{0} [w] = 0$ . So $w \equiv 0$ on every Cauchy slice of $[0, t]$ , hence $w \equiv 0$ on $J^{+} (Σ)$ . Symmetric argument on $J^{-} (Σ)$ gives $w \equiv 0$ on all of $M$ . So $u_{1} = u_{2}$ .

Step 3: Local existence. Near any point $p \in M$ choose a relatively compact globally hyperbolic open neighbourhood $U$ with a Cauchy hypersurface $Σ^{'} \subset U$ containing a piece of $Σ$ . On $U$ the operator $P$ is strictly hyperbolic with constant-coefficient asymptotic behaviour after coordinate rescaling; local existence is the classical result of Leray 1953 / Choquet-Bruhat 1952 / Hörmander 1985 (Ch. XXIII) for the hyperbolic Cauchy problem on a Lorentzian neighbourhood.

Step 4: Global patching. The local solutions on overlapping globally hyperbolic neighbourhoods agree by Step 2 (uniqueness), so they patch to a global solution defined on the union. The Bernal-Sánchez splitting guarantees that the maximal globally hyperbolic neighbourhood through $Σ$ is $M$ itself. The support property $supp (u) \subset J (supp (data) \cup supp (f))$ follows from Step 2 applied to test data that vanish outside the relevant cones.

Step 5: Construction of $E^{\pm}$ . For $f \in C_{c}^{\infty} (M)$ , define $E^{+} f$ as the unique solution of $P u = f$ with zero Cauchy data on a Cauchy hypersurface $Σ$ lying entirely to the past of $supp (f)$ . (Such a $Σ$ exists because $supp (f)$ is compact and the Bernal-Sánchez splitting gives a foliation by Cauchy slices indexed by the time function $t$ .) Uniqueness via Step 2 makes $E^{+}$ well-defined and linear; the support property $supp (E^{+} f) \subset J^{+} (supp f)$ follows because $u$ vanishes on $Σ$ by construction and the energy estimate prevents non-zero values outside the causal future of the source. Symmetric construction with $Σ$ to the future of $supp (f)$ gives $E^{-}$ . Continuity in the appropriate Fréchet topology is a consequence of the continuous-data energy estimate. $■$

Bridge. The Cauchy theorem builds toward 13.09.03, where the Pauli-Jordan commutator $E = E^{-} - E^{+}$ is upgraded to a symplectic form on the phase space of Cauchy data, used to define the CCR algebra of the quantised free Klein-Gordon field, and used to formulate the wave-front-set criterion for Hadamard states (Radzikowski 1996). Appears again in 13.09.04, where the existence of Hadamard states for the quantised Klein-Gordon field on a general globally hyperbolic spacetime is reduced via the FNW deformation argument to the existence of the ground state on an ultrastatic spacetime, with $E$ controlling the propagation of the Hadamard property across the deformation. The pattern — a hyperbolic operator with well-posed Cauchy problem on a globally hyperbolic Lorentzian background — recurs in the Maxwell equation, in the Dirac equation (Dimock 1982 ^{[Dimock 1980]}), in linearised Einstein gravity, and in the Yang-Mills equation, all of which inherit Green's-operator constructions of the same kind. The bridge from Lorentzian geometry to algebraic QFT passes through these propagators: $E$ is the integral kernel of the CCR commutator $[\hat{ϕ} (f), \hat{ϕ} (g)] = i ℏ E (f, g) 1$ , the basic object of the quantised theory.

Exercises Intermediate+

Exercise 1 (easy, symbolic).

Write out the d'Alembertian $□_{g}$ in the Bernal-Sánchez unit-lapse gauge $g = - d t^{2} + h_{t}$ on $M = R \times Σ$ . Express the result in terms of the spatial Laplace-Beltrami operator $Δ_{h_{t}}$ and a time-dependent correction involving the volume element.

Hint

Use the coordinate formula $□_{g} = ∣ g ∣^{- 1/2} \partial_{μ} (∣ g ∣^{1/2} g^{μν} \partial_{ν})$ . In the unit-lapse gauge, $∣ g ∣^{1/2} = det h_{t}$ depends on $t$ .

Answer

In the unit-lapse gauge, $g_{00} = - 1$ , $g_{0 i} = 0$ , $g_{ij} = (h_{t})_{ij}$ , so $∣ g ∣^{1/2} = det h_{t}$ and $g^{00} = - 1$ , $g^{ij} = (h_{t})^{ij}$ . Therefore

□_{g} u = \frac{1}{det h _{t}} \partial_{t} (det h_{t} (- 1) \partial_{t} u) + \frac{1}{det h _{t}} \partial_{i} (det h_{t} (h_{t})^{ij} \partial_{j} u) .

The first term simplifies to $- \partial_{t}^{2} u - (\partial_{t} ln det h_{t}) \partial_{t} u$ . The second is the spatial Laplace-Beltrami operator $Δ_{h_{t}}$ acting on $u$ at fixed $t$ . So

□_{g} u = - \partial_{t}^{2} u - (\partial_{t} ln det h_{t}) \partial_{t} u + Δ_{h_{t}} u .

The first-derivative-in- $t$ term is the friction induced by the time-varying spatial volume; on a stationary spacetime ( $h_{t} = h_{0}$ independent of $t$ ) it vanishes and one recovers the clean wave-equation form $- \partial_{t}^{2} u + Δ_{h_{0}} u = 0$ .

Exercise 2 (easy, multiple choice).

A massive relativistic scalar particle of mass $m$ in flat spacetime has plane-wave solutions $ϕ (t, x) = e^{- iω t + i k \cdot x}$ . Which dispersion relation expresses the on-shell condition?

A. $ω = ∣ k ∣$ B. $ω^{2} = ∣ k ∣^{2}$ C. $ω^{2} = ∣ k ∣^{2} + m^{2}$ D. $ω = m$

Hint

The relativistic energy-momentum relation for a particle of mass $m$ is $E^{2} = p^{2} c^{2} + m^{2} c^{4}$ . In units $c = ℏ = 1$ this is $ω^{2} = ∣ k ∣^{2} + m^{2}$ .

Answer

C. $ω^{2} = ∣ k ∣^{2} + m^{2}$ .

The plane-wave ansatz $ϕ = e^{- iω t + i k \cdot x}$ has time-second-derivative $- ω^{2} ϕ$ and spatial Laplacian $- ∣ k ∣^{2} ϕ$ . The Klein-Gordon equation reduces to an algebraic relation among $ω$ , $k$ and $m$ , with on-shell solutions satisfying $ω^{2} = ∣ k ∣^{2} + m^{2}$ . This is the relativistic energy-momentum relation $E^{2} = p^{2} c^{2} + m^{2} c^{4}$ in natural units, with $E = ℏ ω$ and $p = ℏ∣ k ∣$ . The condition selects a two-sheeted hyperboloid in four-momentum space (positive and negative energy branches), and the choice of the positive branch is the input that distinguishes "particle" from "antiparticle" in the Pauli-Weisskopf second-quantisation construction.

Option A is the dispersion of a massless wave; option B is the algebraic square of A; option D is the rest-energy with no kinetic-energy contribution.

Exercise 3 (medium, symbolic).

Show that the difference $w := u_{1} - u_{2}$ of two smooth solutions of the same Klein-Gordon Cauchy problem $(P u = f, u ∣_{Σ} = u_{0}, n^{μ} \nabla_{μ} u ∣_{Σ} = u_{1})$ on a globally hyperbolic spacetime is identically zero. State explicitly which step uses the energy estimate.

Hint

Set up the energy $E_{t} [w]$ on a slab. Use $P w = 0$ and zero Cauchy data on $Σ$ to get $E_{0} [w] = 0$ .

Answer

The difference $w = u_{1} - u_{2}$ satisfies $P w = P u_{1} - P u_{2} = f - f = 0$ and $w ∣_{Σ} = u_{0} - u_{0} = 0$ , $n^{μ} \nabla_{μ} w ∣_{Σ} = u_{1} - u_{1} = 0$ — so $w$ solves the homogeneous Cauchy problem with zero data on $Σ$ .

The energy on the initial slice is $$ \mathcal{E}0[w] = \int\Sigma \Big(|n^\mu \nabla_\mu w|^2 + h_0^{ij}\partial_i w,\partial_j w + (m^2 + |V|)w^2\Big)\beta, d\mathrm{vol}_{h_0} = 0 $$ because all three integrands vanish by the zero-data condition.

The energy estimate from Step 1 of the Cauchy-problem proof gives, for any $t > 0$ , $$ \mathcal{E}_t[w] \leq C,\Big(\mathcal{E}0[w] + \int_0^t |0|{L^2(\Sigma_s)}^2, ds\Big) = 0. $$ So $E_{t} [w] = 0$ for all $t > 0$ . Since the integrand of $E_{t}$ is non-negative (it is a sum of squares times a positive lapse and a positive measure on a slice), each term must vanish almost everywhere. By smoothness of $w$ , $w \equiv 0$ on $Σ_{t}$ for every $t > 0$ , hence on $J^{+} (Σ)$ . A symmetric argument for $t < 0$ gives $w \equiv 0$ on $J^{-} (Σ) = M ∖ J^{+} (Σ) \cup Σ$ . Combining, $w \equiv 0$ on $M$ .

The step that uses the energy estimate is the bound $E_{t} [w] \leq C E_{0} [w]$ ; without it, the energy on a later slice could in principle exceed the initial energy (which is zero here), and we could not conclude $w \equiv 0$ .

Exercise 4 (medium, numeric).

A solution $u$ of $□ u = 0$ on Minkowski $R^{1, 3}$ has initial data $u (0, \cdot)$ and $\partial_{t} u (0, \cdot)$ supported in the ball $∣ x ∣ \leq 1$ . At time $t = 4$ , the solution $u (4, \cdot)$ is supported in some annular region $r_{-} \leq ∣ x ∣ \leq r_{+}$ . Find the values of $r_{-}$ and $r_{+}$ for generic non-zero initial data, assuming dimension 4 of spacetime (i.e. 3 spatial dimensions).

Hint

For the massless wave equation in $3 + 1$ dimensions, Huygens' principle holds: the solution at a point $(t, x)$ depends only on the initial data at points $x_{0}$ at exactly $∣ x - x_{0} ∣ = t$ , not on closer points. So the support inside the future cone is an annulus, not a ball.

Answer

$r_{-} = 3$ , $r_{+} = 5$ .

In Minkowski $R^{1, 3}$ — three spatial dimensions — the massless scalar wave equation has the strong Huygens principle: the solution at $(t, x)$ depends only on initial data at points $x_{0}$ satisfying $∣ x_{0} - x ∣ = t$ (the boundary of the past light cone), not on points inside. So the support of $u (4, \cdot)$ lies in the spherical shell ${(t, x) : 3 \leq ∣ x ∣ \leq 5}$ : the outer radius $5 = 1 + 4$ from finite propagation, the inner radius $3 = 4 - 1$ from Huygens' principle.

The strong Huygens principle is a feature of the massless wave equation in odd spatial dimensions $\geq 3$ . In two spatial dimensions or for the massive Klein-Gordon equation in any dimension, the wake fills the interior of the future cone and the support inside the cone is a solid ball, not an annulus. For the massless equation in $3 + 1$ dimensions, the Bessel-function term in the propagator vanishes (no wake) and only the $δ$ -on-the-light-cone term remains.

Exercise 5 (medium, symbolic).

Show that the causal propagator $E = E^{-} - E^{+}$ satisfies $P E = 0$ on $C_{c}^{\infty} (M)$ — that is, $E$ maps test functions to solutions of the homogeneous Klein-Gordon equation.

Hint

Apply the defining relation $P E^{\pm} f = f$ to $E^{+}$ and $E^{-}$ separately, and then subtract.

Answer

By definition $P E^{\pm} f = f$ for every $f \in C_{c}^{\infty} (M)$ . Subtracting,

P (E^{-} f) - P (E^{+} f) = f - f = 0.

By linearity of $P$ , $P (E^{-} f - E^{+} f) = P (E f) = 0$ . Since $f$ was arbitrary, $P E = 0$ on $C_{c}^{\infty} (M)$ . Equivalently: $E$ maps any compactly supported smooth source $f$ to a solution $E f \in C^{\infty} (M)$ of the homogeneous Klein-Gordon equation, with support contained in $J (supp f) = J^{+} (supp f) \cup J^{-} (supp f)$ . This is the relation that lets $E$ generate the symplectic phase space of Cauchy data: the image $E (C_{c}^{\infty} (M))$ is precisely the space of smooth homogeneous solutions whose Cauchy data on any Cauchy slice is compactly supported.

Exercise 6 (medium, symbolic).

In Minkowski $R^{1, 1}$ (one spatial dimension), compute the explicit retarded propagator $E^{+} (x)$ for the massless wave equation $□ u = 0$ where $□ = - \partial_{t}^{2} + \partial_{x}^{2}$ . Verify finite propagation speed.

Hint

Use the d'Alembert decomposition $u = f (x - t) + g (x + t)$ . The retarded propagator solves $□ E^{+} = δ (t) δ (x)$ with $supp (E^{+}) \subset {(t, x) : t \geq ∣ x ∣}$ .

Answer

Define $E^{+} (t, x) = \frac{1}{2} θ (t - ∣ x ∣)$ . Then

$E^{+}$ is supported in ${t \geq ∣ x ∣}$ , the closed future light cone of the origin (a wedge in $1 + 1$ dimensions).
$E^{+}$ is the unique distribution with this support satisfying $□ E^{+} = δ (t) δ (x)$ . To check: $□ θ (t - ∣ x ∣) = (- \partial_{t}^{2} + \partial_{x}^{2}) θ (t - ∣ x ∣)$ . Computing the first term: $\partial_{t} θ (t - ∣ x ∣) = δ (t - ∣ x ∣)$ , $\partial_{t}^{2} = δ^{'} (t - ∣ x ∣)$ , so $- \partial_{t}^{2} = - δ^{'} (t - ∣ x ∣)$ . For the second: $\partial_{x} θ (t - ∣ x ∣) = - sgn (x) δ (t - ∣ x ∣)$ and $\partial_{x}^{2} = - 2 δ (t - ∣ x ∣) δ (x) + sgn^{2} (x) δ^{'} (t - ∣ x ∣) = δ^{'} (t - ∣ x ∣)$ — the sign-squared gives 1 for $x \neq = 0$ . Adding, $□ θ (t - ∣ x ∣) = - δ^{'} (t - ∣ x ∣) + δ^{'} (t - ∣ x ∣) + 2 δ (t) δ (x) = 2 δ (t) δ (x)$ . So $□ (\frac{1}{2} θ (t - ∣ x ∣)) = δ (t) δ (x)$ .

Finite propagation: if $f \in C_{c}^{\infty} (R^{1, 1})$ has $supp (f) \subset K$ a compact set in spacetime, then $(E^{+} * f) (t, x) = \int_{K} \frac{1}{2} θ ((t - t^{'}) - ∣ x - x^{'} ∣) f (t^{'}, x^{'}) d t^{'} d x^{'}$ , vanishing unless there exists $(t^{'}, x^{'}) \in K$ with $t - t^{'} \geq ∣ x - x^{'} ∣$ , i.e. $(t, x) \in J^{+} (K)$ . So $supp (E^{+} * f) \subset J^{+} (K)$ .

Note that in $1 + 1$ dimensions, the massless wave equation does not satisfy the strong Huygens principle (which holds only in odd spatial dimensions $\geq 3$ ). The retarded propagator $E^{+}$ is supported on the entire future wedge, not just the future light cone boundary. The wake fills the interior of the cone.

Exercise 7 (hard, symbolic).

Show that the bilinear form $(f, g) \mapsto \int_{M} f (x) (E g) (x) d vol_{g} (x)$ on $C_{c}^{\infty} (M) \times C_{c}^{\infty} (M)$ is antisymmetric, where $E = E^{-} - E^{+}$ is the causal propagator. Conclude that $E$ defines an antisymmetric distributional kernel $E (x, y) = - E (y, x)$ on $M \times M$ .

Hint

Integration by parts twice using $P E^{\pm} = id$ and $E^{\pm} P = id$ on $C_{c}^{\infty} (M)$ . Use the fact that $E^{+}$ and $E^{-}$ are mutually adjoint up to a sign: $\int f E^{+} g = \int g E^{-} f$ for compactly supported $f, g$ .

Answer

First the mutual-adjointness. For $f, g \in C_{c}^{\infty} (M)$ , set $u = E^{+} g$ , so $P u = g$ and $supp (u) \subset J^{+} (supp g)$ . Similarly $v = E^{-} f$ , so $P v = f$ and $supp (v) \subset J^{-} (supp f)$ . Then

\int_{M} f u d vol_{g} = \int_{M} (P v) u d vol_{g} = \int_{M} v (P u) d vol_{g} + boundary,

by integration by parts twice (the wave operator $P = □_{g} + m^{2} + V$ is formally self-adjoint with respect to $d vol_{g}$ ). The boundary term vanishes because $supp (u) \cap supp (v) \subset J^{+} (supp g) \cap J^{-} (supp f)$ is compact (global hyperbolicity!) and the integration is over an open region beyond the supports, where one or the other of $u, v$ vanishes. So

\int_{M} f (E^{+} g) d vol_{g} = \int_{M} (E^{-} f) g d vol_{g} .

This is the relation $E^{+} = (E^{-})^{*}$ where $*$ is the formal adjoint with respect to the natural $L^{2}$ pairing. Now compute:

\int_{M} f (E g) d vol_{g} = \int_{M} f (E^{-} g) d vol_{g} - \int_{M} f (E^{+} g) d vol_{g} .

Apply mutual adjointness $E^{+} = (E^{-})^{*}$ to the second term: $\int f (E^{+} g) = \int (E^{-} f) g = \int g (E^{-} f)$ . The first term: $\int f (E^{-} g) = \int (E^{+} f) g = \int g (E^{+} f)$ . Substituting,

\int_{M} f (E g) = \int_{M} g (E^{+} f) - \int_{M} g (E^{-} f) = - \int_{M} g (E^{-} f - E^{+} f) = - \int_{M} g (E f) .

So $\int f (E g) = - \int g (E f)$ , i.e. $E$ is antisymmetric as a bilinear form. As a distributional kernel on $M \times M$ , this reads $E (x, y) = - E (y, x)$ .

The antisymmetry is what makes $E$ play the role of a symplectic form on the phase space of Cauchy data when restricted to $C_{c}^{\infty} (Σ) \oplus C_{c}^{\infty} (Σ)$ — the foundation of the CCR algebra in 13.09.03.

Exercise 8 (hard, symbolic).

For the massive Klein-Gordon equation $(□ + m^{2}) u = 0$ on Minkowski $R^{1, 3}$ with $m > 0$ , derive the explicit form of the retarded propagator inside the future light cone using the spectral method (Fourier transform in $x$ , contour integration in the conjugate frequency).

Hint

In Fourier-transformed form, $E^{+} = (2 π)^{- 4} \int d^{4} k e^{ik \cdot x} / (k^{2} + m^{2})$ with a contour pushed below both poles $k^{0} = \pm ω_{k}$ at the real axis (Feynman-Stueckelberg pole rule for retarded). Closing the contour in the lower half-plane for $x^{0} > 0$ picks up both poles.

Answer

In four-momentum representation, the Klein-Gordon operator $□ + m^{2}$ in mostly-plus signature has principal symbol $- k_{0}^{2} + ∣ k ∣^{2} + m^{2} = - (k_{0}^{2} - ω_{k}^{2})$ , where $ω_{k} = ∣ k ∣^{2} + m^{2} > 0$ . The Fourier-transformed propagator with retarded pole prescription is

\tilde{E}^{+} (k) = \frac{1}{- ( k _{0} + i ϵ ) ^{2} + ω _{k}^{2}} = \frac{- 1}{( k _{0} - ω _{k} + i ϵ ) ( k _{0} + ω _{k} + i ϵ )} .

(Both poles displaced below the real axis is the retarded prescription.) Inverse-Fourier-transform: $E^{+} (x) = (2 π)^{- 4} \int d^{4} k e^{i (k \cdot x)} \tilde{E}^{+} (k)$ , with $k \cdot x = - k_{0} x^{0} + k \cdot x$ . For $x^{0} > 0$ , close the $k_{0}$ contour in the lower half-plane (the contour where $e^{- i k_{0} x^{0}}$ decays); both poles are picked up by residues, giving

E^{+} (x) = - i \int \frac{d ^{3} k}{( 2 π ) ^{3}} \frac{e ^{i k \cdot x}}{2 ω _{k}} (e^{- i ω_{k} x^{0}} - e^{i ω_{k} x^{0}}) θ (x^{0}) .

This is the standard form of the retarded propagator as a Fourier integral. Using rotational invariance to do the angular $k$ integral first and then the radial integral involves Bessel functions; the final answer is

E^{+} (x) = \frac{1}{2 π} θ (x^{0}) [δ (x^{μ} x_{μ}) - \frac{m}{2 x ^{μ} x _{μ}} θ (x^{μ} x_{μ}) J_{1} (m x^{μ} x_{μ})],

where $x^{μ} x_{μ} = - ∣ x^{0} ∣^{2} + ∣ x ∣^{2}$ (mostly-plus signature, so $- ∣ x^{0} ∣^{2} + ∣ x ∣^{2} \geq 0$ on the future light cone interior, $\leq 0$ on the future light cone exterior). The $δ (x^{μ} x_{μ})$ term sits exactly on the future light cone boundary; the $J_{1}$ term, with the $θ (x^{μ} x_{μ})$ cut-off, sits inside the light cone — the massive wake. In the massless limit $m \to 0$ , the Bessel-function tail vanishes (since $J_{1} (0) = 0$ ) and only the $δ$ -on-the-cone term remains, recovering the strong Huygens principle for massless waves in 3+1.

The detailed computation is the standard quantum-field-theory calculation; see Bär-Ginoux-Pfäffle 2007 ^{[Bär-Ginoux-Pfäffle 2007 §3.4]} for the careful version, or Friedlander 1975 ^{[Friedlander 1975 Ch. 5]} for the original textbook derivation.

Exercise 9 (hard, symbolic).

On a smooth globally hyperbolic spacetime, show that the image $S := E (C_{c}^{\infty} (M))$ of the causal propagator is exactly the space of smooth solutions $u$ of the homogeneous Klein-Gordon equation $(□_{g} + m^{2}) u = 0$ whose Cauchy data $(u ∣_{Σ}, n^{μ} \nabla_{μ} u ∣_{Σ})$ on any Cauchy hypersurface $Σ$ is compactly supported.

Hint

For one inclusion: $u = E f$ for $f \in C_{c}^{\infty} (M)$ has support in $J (supp f)$ , which intersects any Cauchy slice $Σ$ in a compact set. For the reverse inclusion: given a homogeneous solution $u$ with compactly supported Cauchy data on $Σ$ , construct an $f$ with $E f = u$ via a partition-of-unity argument on the slab $J^{-} (Σ) \cap J^{+} (earlier slice)$ .

Answer

( $\subset$ ) Let $f \in C_{c}^{\infty} (M)$ . Then $u = E f = E^{-} f - E^{+} f$ satisfies $P u = P E^{-} f - P E^{+} f = f - f = 0$ (Exercise 5), so $u$ is a homogeneous solution. Its support lies in $J (supp f) = J^{+} (supp f) \cup J^{-} (supp f)$ . Restricting to any Cauchy slice $Σ$ , the intersection $J (supp f) \cap Σ$ is compact (global hyperbolicity: causal future of a compact set is compact when intersected with a Cauchy slice). So the Cauchy data of $u$ on $Σ$ is compactly supported.

( $\supset$ ) Let $u \in C^{\infty} (M)$ with $P u = 0$ and compactly supported Cauchy data on $Σ$ . By the Cauchy-problem uniqueness theorem (Step 2 of the existence proof), $u$ is determined by its data on $Σ$ . The support of $u$ lies in $J (supp (u ∣_{Σ}) \cup supp (n^{μ} \nabla_{μ} u ∣_{Σ})) =: J (K)$ .

Choose a smooth bump function $χ : M \to [0, 1]$ supported in a slab $[t_{-}, t_{+}] \times Σ$ around $Σ$ (in the Bernal-Sánchez splitting), with $χ \equiv 1$ on a thinner slab $[t_{-} /2, t_{+} /2] \times Σ$ . Set $f := P (χ u)$ . By construction $f \in C_{c}^{\infty} (M)$ since $χ u$ has compact spacetime support (inside the slab intersected with $J (K)$ — compact by global hyperbolicity). One checks $E f = u$ : outside the supports, both sides are the unique extension of the homogeneous Cauchy data on $Σ$ , and by uniqueness they agree.

The image $S$ of $E$ is the canonical phase space of the free Klein-Gordon field on $(M, g)$ , equipped with the antisymmetric form $(u, v) \mapsto \int_{M} f_{u} v d vol_{g}$ for any pre-image $f_{u}$ with $E f_{u} = u$ — the form is well-defined on the quotient $C_{c}^{\infty} (M) / P C_{c}^{\infty} (M)$ , isomorphic to $S$ via $E$ .

Lean formalization Intermediate+

Mathlib has no Lorentzian-metric infrastructure, no d'Alembertian on a pseudo-Riemannian manifold, and no theory of hyperbolic Cauchy problems on a curved background as of 2026-05. The closest layers are Geometry.Manifold.SmoothManifoldWithCorners (smooth manifolds), Geometry.Manifold.MetricSpace (positive-definite Riemannian metrics), and the partial distribution theory in Analysis.Distribution. The pseudo-Riemannian-metric-on-a-smooth-manifold structure, the d'Alembertian as a differential operator, the energy estimate over a Cauchy hypersurface, the finite propagation speed property, the advanced and retarded fundamental solutions, the Pauli-Jordan commutator, and the wave-front-set machinery for distributions on a manifold do not yet exist as Mathlib objects.

The Lean 4 SpaceTime project (M. Larson and collaborators, separate from Mathlib) has a partial Lorentzian-metric layer; the wave-equation theory and the Green's-operator constructions are not formalised in any current Lean library. The Hörmander-style microlocal analysis required for the wave-front-set characterisation of $E$ would itself be a substantial formalisation project.

lean_status: none reflects this gap; no Lean module ships with this unit. the Mathlib gap analysis names the specific missing infrastructure. Tyler's review attests Intermediate-tier correctness of the Cauchy-problem statement, the energy estimate framework, and the Minkowski worked example. The Master-tier microlocal characterisation of $E$ (the bridge to Radzikowski's Hadamard-state criterion in 13.09.03) is flagged for external review by a Lorentzian-PDE / algebraic-QFT specialist.

Advanced results Master

Three structural developments extend the basic Cauchy-problem statement to the depth Gérard 2019 Ch. 5-6 requires for the algebraic QFT programme on curved spacetimes.

Sobolev well-posedness and the energy norm. The smooth Cauchy-problem statement extends to a Sobolev-space well-posedness theorem ^{[Hörmander 1985 §23.2]}. For $s \in R$ and initial data $(u_{0}, u_{1}) \in H^{s + 1} (Σ) \oplus H^{s} (Σ)$ with the source $f \in L_{loc}^{1} ([t_{-}, t_{+}]; H^{s} (Σ_{t}))$ , the Cauchy problem has a unique solution $u \in C^{0} ([t_{-}, t_{+}]; H^{s + 1} (Σ_{t})) \cap C^{1} ([t_{-}, t_{+}]; H^{s} (Σ_{t}))$ with the energy estimate $∥ u (t) ∥_{H^{s + 1} (Σ_{t})} + ∥ \partial_{t} u (t) ∥_{H^{s} (Σ_{t})} \leq C (∥ u_{0} ∥_{H^{s + 1}} + ∥ u_{1} ∥_{H^{s}} + \int ∥ f ∥_{H^{s}} d t)$ . The compactly-supported $C^{\infty}$ statement of the Intermediate tier is the $s \to \infty$ limit, but the Sobolev statement is what underlies the operator-theoretic CCR construction of 13.09.03, where the one-particle Hilbert space is built from the $H^{1} \oplus L^{2}$ phase space of Cauchy data, and the Hadamard parametrix subtraction of 13.09.06 requires precise control of the scaling degree of $E$ near the diagonal.

Symplectic structure on the phase space of Cauchy data. The antisymmetric bilinear form $σ : S \times S \to R$ given by $σ (u, v) := \int_{M} f_{u} v d vol_{g}$ (for $u = E f_{u}$ , $v = E f_{v}$ ) descends to a well-defined non-degenerate symplectic form on the quotient $S = C_{c}^{\infty} (M) / P C_{c}^{\infty} (M)$ , identified via $E$ with the space of homogeneous solutions with compactly supported Cauchy data. Equivalently, on the Cauchy-data space $D_{Σ} := C_{c}^{\infty} (Σ) \oplus C_{c}^{\infty} (Σ)$ , the symplectic form pulls back to

σ_{Σ} ((u_{0}, u_{1}), (v_{0}, v_{1})) = \int_{Σ} (u_{1} v_{0} - u_{0} v_{1}) d vol_{h_{Σ}},

a form Cauchy-independent of $Σ$ : choosing a different Cauchy hypersurface $Σ^{'}$ and pulling back $σ$ along the unique-extension map $D_{Σ} \to D_{Σ^{'}}$ yields the same symplectic form on the abstract phase space ^{[Wald 1994 §3.3]}. The independence of $Σ$ is the geometric realisation of the conservation law associated to the Klein-Gordon Lagrangian by Noether's theorem; analytically it is the divergence-theorem statement $\int_{Σ} j^{μ} n_{μ} = \int_{Σ^{'}} j^{μ} n_{μ}$ for the symplectic current $j^{μ} = u \nabla^{μ} v$ .

Wave-front set of the causal propagator. The deepest structural result on $E$ for the curved-spacetime QFT programme is its wave-front-set characterisation, proved in Minkowski by Hörmander in the late 1960s ^{[Hörmander 1985 §23.4]} and used by Duistermaat-Hörmander 1972 Acta Math. 128 to establish propagation of singularities for normally hyperbolic operators. On Minkowski $R^{1, 3}$ the wave-front set of the causal propagator $E$ is

WF (E) = {((x, ξ), (y, - η)) \in T^{*} R^{1, 3} \times T^{*} R^{1, 3} : (x - y) is null, ξ is conormal to the null line through x, y, η is the parallel transport of ξ} .

The conormality condition means $ξ$ is a null covector pointing along the light cone connecting $x$ and $y$ . On a general globally hyperbolic spacetime the same structure holds with "null line" replaced by "null geodesic" and "parallel transport" replaced by the parallel transport along the connecting geodesic ^{[Radzikowski 1996]}. The wave-front set of $E$ is the light-cone bicharacteristic relation on $T^{*} M ∖ 0$ — the set of pairs of cotangent vectors related by Hamiltonian flow of the principal symbol $g^{μν} ξ_{μ} ξ_{ν}$ along null directions.

This characterisation is the load-bearing input to Radzikowski's wave-front-set criterion for Hadamard states in 13.09.03: a quasi-free state $ω$ on the CCR algebra is Hadamard iff the two-point function $ω_{2} \in D^{'} (M \times M)$ has wave-front set equal to the subset of the bicharacteristic relation where $ξ$ is future-directed at $x$ . The criterion separates the "positive-frequency part" of the wave-front set from the "negative-frequency part" by direction of $ξ$ ; the causal propagator $E$ provides the full bicharacteristic relation, and a Hadamard state is the projection onto the future-directed half.

Synthesis. The Klein-Gordon Cauchy problem on a globally hyperbolic background is the analytic shadow of the geometric condition from 13.09.01: compactness of $J^{+} (p) \cap J^{-} (q)$ together with the smooth Bernal-Sánchez splitting makes $□_{g} + m^{2}$ a strictly hyperbolic operator with well-posed Cauchy problem and finite-speed-of-propagation Green's functions $E^{\pm}$ . The Pauli-Jordan commutator $E = E^{-} - E^{+}$ encodes the symplectic structure of the classical phase space, the CCR commutator of the quantised theory, and (via its wave-front set) the bicharacteristic relation that defines Hadamard states on a curved background. This structure builds toward 13.09.03, where the CCR algebra and Radzikowski's wave-front-set criterion are stated; appears again in 13.09.04, where the existence of Hadamard states on every globally hyperbolic spacetime is established via the FNW deformation argument with $E$ controlling the propagation of the Hadamard property; and recurs throughout the renormalisation programme of 13.09.06, where the Hadamard parametrix subtraction defines Wick polynomials in terms of the singular structure of $E$ . The bridge from Lorentzian geometry to algebraic QFT passes through this Cauchy problem: without it, no propagators, no symplectic phase space, no CCR algebra, no Hadamard states, no perturbative QFT in curved spacetime.

Full proof set Master

Proposition (Uniqueness of the Cauchy problem). Let $(M, g)$ be globally hyperbolic with Cauchy hypersurface $Σ$ , and let $P = □_{g} + m^{2} + V$ with $V \in C^{\infty} (M)$ and $m \geq 0$ . If $u \in C^{\infty} (M)$ satisfies $P u = 0$ , $u ∣_{Σ} = 0$ , and $n^{μ} \nabla_{μ} u ∣_{Σ} = 0$ , then $u \equiv 0$ on $M$ .

Proof. Work in the Bernal-Sánchez splitting $g = - β^{2} d t^{2} + h_{t}$ with $Σ = Σ_{0} = {t = 0}$ . Define the energy on $Σ_{t}$ by $E_{t} [u] = \int_{Σ_{t}} (∣ \partial_{t} u ∣^{2} + h_{t}^{ij} \partial_{i} u \partial_{j} u + (m^{2} + ∣ V ∣) u^{2}) β d vol_{h_{t}}$ . Differentiation in $t$ under the integral sign, using $P u = 0$ and the divergence theorem on the slab $[0, t] \times Σ$ , gives $\partial_{t} E_{t} [u] \leq C E_{t} [u]$ for a constant $C$ depending on the geometry but not on $u$ (the cross-term $\partial_{t} ∣ \partial_{t} u ∣^{2}$ produces a $□_{g} u \cdot \partial_{t} u$ contribution that vanishes by $P u = 0$ up to a manageable $V u \partial_{t} u$ remainder bounded by $E_{t}$ ). Grönwall's inequality then yields $E_{t} [u] \leq e^{C t} E_{0} [u]$ .

By the hypotheses $u ∣_{Σ} = 0$ , $\partial_{t} u ∣_{Σ} = 0$ , all integrands in $E_{0} [u]$ vanish, so $E_{0} [u] = 0$ . Therefore $E_{t} [u] = 0$ for all $t \geq 0$ . Each integrand of $E_{t}$ is non-negative; non-negativity plus zero integral plus smoothness of $u$ forces $u = 0$ , $\partial_{t} u = 0$ , $\partial_{i} u = 0$ at every point of $Σ_{t}$ . So $u \equiv 0$ on $J^{+} (Σ)$ . A time-reversed argument gives $u \equiv 0$ on $J^{-} (Σ)$ . Combining, and using $M = J^{+} (Σ) \cup J^{-} (Σ)$ from 13.09.01, $u \equiv 0$ on $M$ . $□$

Proposition (Support property of $E^{\pm}$ ). For every $f \in C_{c}^{\infty} (M)$ , $supp (E^{\pm} f) \subset J^{\pm} (supp f)$ .

Proof. Take $E^{+}$ ; the argument for $E^{-}$ is symmetric. By construction $E^{+} f$ is the unique solution of $P u = f$ with $u ∣_{Σ} = 0$ and $\partial_{t} u ∣_{Σ} = 0$ on a Cauchy hypersurface $Σ$ entirely to the past of $supp (f)$ . Suppose $p \in supp (E^{+} f)$ . Then there is a sequence of points $p_{n} \to p$ with $(E^{+} f) (p_{n}) \neq = 0$ . If $p \in / J^{+} (supp f)$ , then for some open neighbourhood $U$ of $p$ , no future-directed causal curve from $supp (f)$ reaches $U$ . Choose a Cauchy hypersurface $Σ^{'}$ entirely in $J^{-} (p) ∖ J^{+} (supp f)$ — possible because $p \in / J^{+} (supp f)$ . On $Σ^{'}$ both $E^{+} f$ and $\partial_{t} (E^{+} f)$ vanish, because $Σ^{'}$ lies to the past of $supp (f)$ and $E^{+} f$ inherits the zero-Cauchy-data property from $Σ$ through the slab below $supp (f)$ . Apply the uniqueness proposition to the homogeneous part of the equation between $Σ^{'}$ and a slice through $p$ : $E^{+} f$ must vanish in this slab. So $p \in / supp (E^{+} f)$ , contradiction. Hence $supp (E^{+} f) \subset J^{+} (supp f)$ . $□$

Proposition (Mutual adjointness $E^{+} = (E^{-})^{*}$ and antisymmetry of $E$ ). For $f, g \in C_{c}^{\infty} (M)$ , $\int_{M} f (E^{+} g) d vol_{g} = \int_{M} g (E^{-} f) d vol_{g}$ , and the causal propagator $E = E^{-} - E^{+}$ is antisymmetric: $E (x, y) = - E (y, x)$ as a distributional kernel.

Proof. Exercise 7. The key step is integration by parts twice: $P$ is formally self-adjoint with respect to $d vol_{g}$ , so $\int u (P v) = \int (P u) v$ for $u, v$ with overlapping compact supports. The boundary term vanishes because $supp (E^{+} g) \cap supp (E^{-} f) \subset J^{+} (supp g) \cap J^{-} (supp f)$ is compact in $M$ by global hyperbolicity. $□$

Proposition (Image of $E$ is the canonical phase space). The map $E : C_{c}^{\infty} (M) \to C^{\infty} (M)$ has kernel $P C_{c}^{\infty} (M)$ and image $S := {u \in C^{\infty} (M) : P u = 0, Cauchy data on every Cauchy slice is compactly supported}$ .

Proof. Exercise 9. The kernel computation: $E f = 0$ iff $E^{+} f = E^{-} f$ . Both sides have disjoint supports for generic $f$ (one in $J^{+}$ , the other in $J^{-}$ ), so they can agree only if both are identically zero, i.e. $f = 0$ as a homogeneous-equation source — equivalently, $f \in P C_{c}^{\infty} (M)$ . The image computation uses the bump-function construction sketched in Exercise 9. $□$

Connections Master

Globally hyperbolic Lorentzian manifolds 13.09.01 supplies the geometric prerequisite: the Cauchy hypersurface $Σ$ , the smooth Bernal-Sánchez splitting $g = - β^{2} d t^{2} + h_{t}$ , and the compactness of causal diamonds $J^{+} (p) \cap J^{-} (q)$ that powers the finite-speed-of-propagation argument. Without global hyperbolicity the Klein-Gordon Cauchy problem fails to be well-posed; the present unit is the analytic content of the geometric condition.
Tensors on smooth manifolds 13.02.01 supplies the differential-geometric language: the metric $g^{μν}$ , the Levi-Civita connection $\nabla$ , the d'Alembertian $□_{g} = g^{μν} \nabla_{μ} \nabla_{ν}$ , the natural volume form $d vol_{g} = ∣ g ∣ d^{4} x$ , and the formal-adjoint computation that makes $P = □_{g} + m^{2} + V$ symmetric with respect to the natural $L^{2}$ pairing.
Wave-front set of a distribution 02.14.01 is the microlocal tool used in the Master-tier characterisation of $WF (E)$ . The wave-front set localises the singularities of $E$ in cotangent directions, and the resulting bicharacteristic relation is exactly the input to Radzikowski's Hadamard-state criterion in 13.09.03. Propagation of singularities along null geodesics is the Hamiltonian-flow analytic statement (Duistermaat-Hörmander 1972).
Schwarzschild solution 13.05.01 and FLRW cosmology 13.08.01 are the canonical curved backgrounds on which the Klein-Gordon equation has been studied. On Schwarzschild the Cauchy problem for a real scalar field is the standard setting for the Hawking radiation calculation (Hawking 1975); on FLRW it underlies the inflationary primordial-perturbation programme. In both cases the Bernal-Sánchez foliation is realised by an explicit choice of time coordinate (Schwarzschild $t$ outside the horizon; cosmic time $t$ in FLRW), and the Klein-Gordon mode expansion reduces to a one-parameter family of ODEs in the spatial Laplace eigenmode.
Wightman axioms 08.10.07 state the flat-space Klein-Gordon quantum field as an operator-valued tempered distribution satisfying Poincaré covariance, microcausality, the spectrum condition, vacuum cyclicity, and cluster decomposition. The microcausality axiom $[\hat{ϕ} (f), \hat{ϕ} (g)] = 0$ for spacelike-separated $f, g$ is exactly the support property of the causal propagator $E$ on Minkowski: $E$ vanishes outside the light cone, so smearing against test functions with spacelike-separated supports gives zero commutator. The present unit's curved-spacetime $E$ is the natural generalisation; the Wightman-axiom framework on a curved background is the BFV 2003 locally-covariant-functor formulation.
Hadamard states via wave-front-set criterion [13.09.03, pending] is the immediate downstream unit. The Cauchy-data symplectic form $σ_{Σ}$ defined here is the basic input to the CCR algebra; the bicharacteristic-relation wave-front set of $E$ is the basic input to Radzikowski's definition of Hadamard states; the two-point function $ω_{2}$ of a Hadamard state has wave-front set equal to the "positive-frequency half" of $WF (E)$ .
Existence of Hadamard states (FNW deformation) [13.09.04, pending] uses the present unit's Green's operators $E^{\pm}$ to control the propagation of the Hadamard property under a globally-hyperbolic deformation of the spacetime. Pulling back the ultrastatic ground state along the deformation map preserves the Hadamard wave-front-set structure because $E^{\pm}$ are deformation-invariant up to a correction with smooth kernel, which the wave-front set ignores.
Bosonic Fock space and second quantisation 08.10.01 is the Hilbert-space-side construction of the free Klein-Gordon QFT on Minkowski. The one-particle Hilbert space is the $L^{2}$ -completion of the Cauchy-data phase space $D_{Σ}$ with respect to the inner product determined by the positive-frequency boundary condition; the symplectic form here becomes the imaginary part of the inner product. On a curved background the role of "positive-frequency" is played by a Hadamard state, and the Fock space is the GNS representation of the CCR algebra in that state.
Wave equation on Minkowski (electromagnetism context, 11.x) is the classical PDE precursor. The Maxwell equations in vacuum reduce to wave equations on the components of the gauge potential after gauge-fixing; the Cauchy problem and the retarded/advanced propagators of electromagnetism are the photon-spin-1 analogue of the present scalar-spin-0 case. The geometric Cauchy-problem theorem extends to all normally hyperbolic operators on a globally hyperbolic background (Bär-Ginoux-Pfäffle 2007 §3).

Historical & philosophical context Master

The Klein-Gordon equation traces back to three independent 1926 derivations: Oskar Klein in Zeitschrift für Physik 37 (1926) 895, Walter Gordon in Zeitschrift für Physik 40 (1926) 117, and Vladimir Fock in Zeitschrift für Physik 39 (1926) 226, each writing down the relativistic generalisation of the Schrödinger equation for a charged scalar particle. The single-particle interpretation faced the difficulty that the conserved current is not positive-definite, producing negative-probability solutions for negative-energy modes. Pauli and Weisskopf in Helvetica Physica Acta 7 (1934) 709 resolved this by reinterpreting the negative-energy solutions as antiparticles, treating $ϕ$ as a quantum field rather than a wave function and second-quantising on the bosonic Fock space — the construction is the prototype for every subsequent relativistic QFT.

The classical Cauchy-problem treatment on a Lorentzian manifold matured in the 1950s. Jean Leray in 1953 ^{[Leray 1953]}, in mimeographed lecture notes at the Institute for Advanced Study in Princeton, introduced the systematic notion of a causal domain and proved local well-posedness of hyperbolic systems on a Lorentzian background; Yvonne Choquet-Bruhat in Acta Math. 88 (1952) 141 ^{[Choquet-Bruhat 1952]} used the same framework to establish local existence and uniqueness of solutions to the vacuum Einstein equations, opening the modern PDE theory of general relativity. André Lichnerowicz in Publ. IHÉS 10 (1961) 5 ^{[Lichnerowicz 1961]} constructed the advanced and retarded propagators for the Klein-Gordon equation on a globally hyperbolic Lorentzian manifold and the Pauli-Jordan commutator $E = E^{-} - E^{+}$ , working in the conventions still standard today.

The QFT-side picture took shape through the work of F. G. Friedlander in his 1975 monograph The Wave Equation on a Curved Space-Time ^{[Friedlander 1975]}, which gave the textbook treatment of the Hadamard parametrix construction for the retarded propagator on a general globally hyperbolic background. Jonathan Dimock in Trans. AMS 269 (1982) 133 ^{[Dimock 1980]} gave the first systematic algebraic-quantisation treatment of the Klein-Gordon and Dirac fields on a curved spacetime, constructing $E^{\pm}$ as Green's operators and using $E$ as the kernel of the CCR commutator. Robert Wald in his 1994 monograph Quantum Field Theory in Curved Spacetime and Black Hole Thermodynamics ^{[Wald 1994]} gave the canonical physicist-side framing, presenting the well-posedness theorem, the symplectic structure on Cauchy data, and the GNS construction of Fock representations from quasi-free states.

The microlocal-analytic reframing of the entire subject came with Marek Radzikowski in Comm. Math. Phys. 179 (1996) 529 ^{[Radzikowski 1996]}, who proved that the Kay-Wald 1991 Hadamard condition on a state is equivalent to a wave-front-set condition on the two-point function expressed in terms of the bicharacteristic relation of $□_{g} + m^{2}$ — the relation whose existence is encoded by the causal propagator $E$ of the present unit. The textbook consolidation of this microlocal paradigm is Christian Gérard in his 2019 monograph Microlocal Analysis of Quantum Fields on Curved Spacetimes ^{[Gérard 2019]} and the parallel classical-PDE treatment of Christian Bär, Nicolas Ginoux and Frank Pfäffle in their 2007 monograph Wave Equations on Lorentzian Manifolds and Quantization ^{[Bär-Ginoux-Pfäffle 2007]}, the latter with a freely-available arXiv version (0806.1036) that has become the canonical free reference for the Cauchy-problem and Green's-operator theory underpinning curved-spacetime QFT.

Bibliography Master

Foundational originator papers:

Klein, O., "Quantentheorie und fünfdimensionale Relativitätstheorie", Z. Phys. 37 (1926), 895-906. [Originating derivation of the Klein-Gordon equation.]
Gordon, W., "Der Comptoneffekt nach der Schrödingerschen Theorie", Z. Phys. 40 (1926), 117-133. [Independent derivation of the Klein-Gordon equation, in the context of the Compton effect.]
Fock, V., "Über die invariante Form der Wellen- und Bewegungsgleichungen für einen geladenen Massenpunkt", Z. Phys. 39 (1926), 226-232.
Pauli, W. & Weisskopf, V., "Über die Quantisierung der skalaren relativistischen Wellengleichung", Helv. Phys. Acta 7 (1934), 709-731. [Second-quantisation reinterpretation that resolved the negative-probability problem of the single-particle Klein-Gordon equation.]
Jordan, P. & Pauli, W., "Zur Quantenelektrodynamik ladungsfreier Felder", Z. Phys. 47 (1928), 151-173. [Originating paper introducing the commutator function $E = E^{-} - E^{+}$ that bears their names.]

Cauchy-problem theory on Lorentzian manifolds:

Leray, J., Hyperbolic differential equations, Institute for Advanced Study, Princeton (1953, mimeographed lecture notes). [Originating systematic treatment of hyperbolic systems on a Lorentzian manifold; introduction of the causal-domain methodology.]
Choquet-Bruhat, Y., "Théorème d'existence pour certains systèmes d'équations aux dérivées partielles non linéaires", Acta Math. 88 (1952), 141-225. [Local well-posedness of the vacuum Einstein equations on Lorentzian neighbourhoods; foundational to the modern PDE theory of GR.]
Lichnerowicz, A., "Propagateurs et commutateurs en relativité générale", Publ. IHÉS 10 (1961), 5-56. [Construction of advanced and retarded propagators on a globally hyperbolic Lorentzian manifold.]
Hadamard, J., Lectures on Cauchy's Problem in Linear Partial Differential Equations (Yale, 1923). [Originating Hadamard parametrix construction for the wave equation; foundational to Friedlander's 1975 textbook treatment.]

Modern monograph treatments:

Friedlander, F. G., The Wave Equation on a Curved Space-Time (Cambridge University Press, 1975). [Classical-PDE textbook treatment of the Hadamard parametrix for the retarded propagator on a globally hyperbolic background.]
Bär, C., Ginoux, N. & Pfäffle, F., Wave Equations on Lorentzian Manifolds and Quantization, ESI Lectures in Mathematics and Physics (EMS, 2007). [Free PDF at arXiv:0806.1036; the canonical modern textbook treatment of the Cauchy problem and Green's operators on a globally hyperbolic Lorentzian manifold.]
Gérard, C., Microlocal Analysis of Quantum Fields on Curved Spacetimes, ESI Lectures in Mathematics and Physics (EMS, 2019). [Modern textbook entry to microlocal-analytic curved-spacetime QFT; Ch. 5-6 cover the Cauchy problem and the causal propagator with the microlocal viewpoint that prepares the wave-front-set criterion for Hadamard states.]
Hörmander, L., The Analysis of Linear Partial Differential Operators III: Pseudo-Differential Operators (Springer, 1985). [Canonical reference for the microlocal-analysis machinery applied to hyperbolic operators; Ch. XXIII covers the Cauchy problem and propagation of singularities for strictly hyperbolic operators on a manifold.]
Wald, R. M., Quantum Field Theory in Curved Spacetime and Black Hole Thermodynamics (University of Chicago Press, 1994). [Physicist-side canonical reference; Ch. 3 treats the Klein-Gordon equation on a globally hyperbolic background, the symplectic structure on Cauchy data, and the construction of the GNS representation for a quasi-free state.]

Algebraic-quantisation programme:

Dimock, J., "Algebras of local observables on a manifold", Comm. Math. Phys. 77 (1980), 219-228. [Originating algebraic-quantisation treatment of the Klein-Gordon field on a curved background.]
Dimock, J., "Dirac quantum fields on a manifold", Trans. AMS 269 (1982), 133-147. [Companion paper treating the Dirac field; the Klein-Gordon Cauchy-problem material is also covered.]
Kay, B. S. & Wald, R. M., "Theorems on the uniqueness and thermal properties of stationary, nonsingular, quasifree states on spacetimes with a bifurcate Killing horizon", Phys. Rep. 207 (1991), 49-136. [Originating Hadamard-form definition of admissible quasi-free states; predecessor to Radzikowski's wave-front-set criterion.]
Radzikowski, M. J., "Micro-local approach to the Hadamard condition in quantum field theory on curved space-time", Comm. Math. Phys. 179 (1996), 529-553. [Wave-front-set characterisation of Hadamard states; bridge from the present unit's causal propagator to the modern microlocal definition.]

Reviews and surveys:

Hollands, S. & Wald, R. M., "Quantum fields in curved spacetime", Phys. Rep. 574 (2015), 1-35. [Free preprint at arXiv:1401.2026; modern review of the field, with extensive treatment of the Cauchy problem and Green's operators on a globally hyperbolic background.]
Brunetti, R., Fredenhagen, K. & Verch, R., "The generally covariant locality principle — a new paradigm for local quantum field theory", Comm. Math. Phys. 237 (2003), 31-68. [Locally-covariant-functor formulation of curved-spacetime QFT; the present unit's well-posedness and Green's-operator construction is the basic input to the BFV functor from globally hyperbolic spacetimes to $C^{*}$ -algebras.]

Prerequisites

13.09.01
13.02.01
02.14.01

Tier anchors

beginner: Wald, General Relativity (1984), Ch. 10 (the wave equation in curved spacetime); Hawking and Ellis, The Large Scale Structure of Space-Time (Cambridge, 1973), §7.4 (selected passages on hyperbolic equations)
intermediate: Bär, Ginoux & Pfäffle, Wave Equations on Lorentzian Manifolds and Quantization (EMS, 2007), Ch. 3 (free PDF at arXiv:0806.1036); Dimock, Trans. AMS 269 (1980) 133; Gérard, Microlocal Analysis of Quantum Fields on Curved Spacetimes (EMS, 2019), §5.3
master: Gérard, Microlocal Analysis of Quantum Fields on Curved Spacetimes (EMS, 2019), Ch. 5-6; Bär-Ginoux-Pfäffle (EMS, 2007), Ch. 3; Wald, Quantum Field Theory in Curved Spacetime and Black Hole Thermodynamics (Chicago, 1994), Ch. 3; Hörmander, The Analysis of Linear Partial Differential Operators III (Springer, 1985), Ch. XXIII

References

Gérard, C. — Microlocal Analysis of Quantum Fields on Curved Spacetimes (EMS Press, 2019) · Ch. 5 (Klein-Gordon equation on a Lorentzian manifold); §5.3 (Cauchy problem); §5.4 (advanced and retarded propagators); Ch. 6 (causal propagator and CCR algebra)
Bär, C., Ginoux, N. & Pfäffle, F. — Wave Equations on Lorentzian Manifolds and Quantization (EMS, 2007) · Ch. 3 (Cauchy problem for normally hyperbolic operators on globally hyperbolic spacetimes); §3.1 (uniqueness and finite propagation speed); §3.2 (existence via Hadamard parametrix and global patching); §3.4 (Green's operators); free PDF at arXiv:0806.1036
Wald, R. M. — Quantum Field Theory in Curved Spacetime and Black Hole Thermodynamics (Chicago, 1994) · Ch. 3 (linear quantum fields on a globally hyperbolic spacetime); §3.2 (Klein-Gordon equation, energy norm, well-posedness); §3.3 (advanced and retarded fundamental solutions and the symplectic structure on the space of solutions)
Hörmander, L. — The Analysis of Linear Partial Differential Operators III: Pseudo-Differential Operators (Springer, 1985) · Ch. XXIII (hyperbolic operators); §23.2 (energy estimates and Cauchy problem for strictly hyperbolic equations); §23.4 (propagation of singularities, Lorentzian-flow form)
Dimock, J. — 'Algebras of local observables on a manifold', Comm. Math. Phys. 77 (1980) 219; 'Dirac quantum fields on a manifold', Trans. AMS 269 (1982) 133 · Klein-Gordon and Dirac quantisation on a globally hyperbolic spacetime via the algebraic / CCR programme; explicit construction of E^± as Green's operators
Leray, J. — 'Hyperbolic differential equations', Institute for Advanced Study, Princeton (1953, mimeographed lecture notes) · Foundational construction of the Cauchy problem for hyperbolic systems on a Lorentzian manifold; introduction of causal-domain methodology
Choquet-Bruhat, Y. — 'Théorème d'existence pour certains systèmes d'équations aux dérivées partielles non linéaires', Acta Math. 88 (1952) 141 · Local well-posedness for hyperbolic systems on a Lorentzian manifold; precursor to global well-posedness on globally hyperbolic backgrounds
Lichnerowicz, A. — 'Propagateurs et commutateurs en relativité générale', Publ. IHÉS 10 (1961) 5 · Construction of advanced/retarded propagators on a globally hyperbolic Lorentzian manifold and the Pauli-Jordan commutator E = E^- - E^+
Hawking, S. W. & Ellis, G. F. R. — The Large Scale Structure of Space-Time (Cambridge, 1973) · §7.4 (the wave equation in curved spacetime, causal propagator)
Friedlander, F. G. — The Wave Equation on a Curved Space-Time (Cambridge, 1975) · Ch. 5-6 (Hadamard parametrix construction for the retarded fundamental solution); the standard classical reference
Radzikowski, M. J. — 'Micro-local approach to the Hadamard condition in quantum field theory on curved space-time', Comm. Math. Phys. 179 (1996) 529 · Wave-front set of the causal propagator E on Minkowski and on a globally hyperbolic spacetime; the bridge to Hadamard states in 13.09.03

Estimated time

beginner: 18m
intermediate: 42m
master: 65m