Hilbert space of differentials; orthogonal decomposition on an open surface

Intuition Beginner

Picture a flow on a surface — wind blowing across it, recorded at every point as a little arrow. Mathematicians package such a flow as a differential: a rule that assigns directional information to each point. On an open surface — one that runs off to infinity or has been punctured, so it is not closed up like a sphere or a doughnut — there is a lot of room, and a natural way to measure the size of a flow is to add up the squared lengths of all its arrows. Flows of finite total size form the working space; we call them the square-integrable, or "finite-energy", differentials.

The central idea is that any finite-energy flow breaks into clean pieces. One piece is a harmonic flow: smooth, source-free and swirl-free, the kind of steady pattern that water settles into. The remaining pieces are "pure gradient" patterns built from a height function, and their swirling counterparts. So the whole space of flows splits into a harmonic part, a gradient part, and a co-gradient part — three perpendicular directions, like decomposing a vector in space into its $x$, $y$, and $z$ components.

This three-way split is the engine behind existence theorems on open surfaces. You want to build a special harmonic flow with prescribed behaviour; the split tells you it sits inside one perpendicular slot, and you reach it by dropping a perpendicular onto that slot. The same dropping-a-perpendicular move that finds the closest point on a plane finds the harmonic flow you need.

Visual Beginner

A schematic of an open surface — drawn as a plane with one puncture and an arrow trailing off toward infinity, signalling that it is not closed. Beside it, a single finite-energy flow is shown being separated into three stacked layers: a smooth swirl-free-and-source-free "harmonic" layer, a "gradient" layer drawn as arrows pointing straight downhill from a shaded height function with a small bump (so the bump is concentrated in one region, not spread to infinity), and a "co-gradient" layer drawn as the same arrows turned ninety degrees. Three short perpendicular axes in the corner label the three layers as mutually at right angles.

Worked example Beginner

Take the simplest open surface: the whole flat plane, the complex numbers, with the usual way of measuring length. Consider the steady horizontal flow whose arrow at every point is one unit long and points to the right. Its squared length is $1$ at each point, and summing $1$ over the entire infinite plane gives infinity — so this flow has infinite energy and is not in our working space.

Now cut the flow down: keep it equal to the rightward unit arrow inside a disc of radius $10$, and smoothly taper it to zero outside radius $20$. The squared length is at most $1$, and it is non-zero only inside a disc of radius $20$, whose area is $400$ times $\pi$, about $1257$. So the total energy is below about $1257$ — a finite number. This tapered flow is in the working space.

Which layer is it in? Inside radius $10$ the flow is the gradient of the height function "$x$-coordinate", a perfect downhill-arrow pattern, so it leans toward the gradient layer. But the tapering region near radius $20$ adds swirl, so it is not purely any one layer; the three-way split would record a definite amount in each. What this tells us: on an open surface, honest finite-energy flows must fade out near infinity, and a single flow can carry components in all three perpendicular layers at once. The decomposition is what measures how much of each it carries.

Check your understanding Beginner

Formal definition Intermediate+

Let $X$ be a Riemann surface, not assumed compact, with a conformal structure. A first-order differential is a real-valued smooth $1$-form $\omega$; in a local holomorphic coordinate $z=x+iy$ it reads $\omega =adx+bdy$ with $a,b$ smooth and real. The conformal Hodge star $\ast$ acts on $1$-forms by $$ *(a, dx + b, dy) = -,b, dx + a, dy, \qquad {} = -1 \text{ on } 1\text{-forms}, $$ a rotation by ninety degrees in each cotangent plane. In real dimension two the star on $1$-forms depends only on the conformal class of the metric, so it is intrinsic to the Riemann surface; this conformal invariance is recorded in 06.11.01 and is the reason the construction needs no metric beyond the complex structure.

The pointwise inner product of $1$-forms is $⟨\omega ,\theta ⟩dA=\omega \wedge \ast \theta$, where $dA$ is the area element. For complex-valued differentials write $\overline{\ast \theta }$ for the conjugate. The Dirichlet inner product is $$ (\omega, \theta) ;=; \iint_X \omega \wedge \overline{\theta}, $$ and a differential is square-integrable when $(\omega, \omega) = \iint_X \omega \wedge \overline{\omega} < \infty$.Let$\Gamma = \Gamma(X)$denotetherealHilbertspaceobtainedbycompletingthesmoothsquare-integrabledifferentialsinthenorm$|\omega| = (\omega, \omega)^{1/2}$.Completionisneededbecauselimitsoffinite-energydifferentialsneedonlybe$L^2$, not smooth.

Three closed subspaces organise $\Gamma$. Write $d$ for the exterior derivative and recall that $\omega$ is closed when $d\omega =0$ and co-closed when $d\ast \omega =0$.

• ${\Gamma }_h=\{\omega \in \Gamma :d\omega =0\text{ and }d\ast \omega =0\text{ weakly}\}$, the harmonic differentials — closed and co-closed.
• ${\Gamma }_{e0}=\overline{\{df:f\in C_c^{\infty }(X)\}}$, the $L^2$-closure of differentials of compactly-supported smooth functions.
• ${\Gamma }_{e0}^{\ast }=\ast {\Gamma }_{e0}=\overline{\{\ast df:f\in C_c^{\infty }(X)\}}$, its image under the star.

A differential is exact if $\omega =df$ globally and co-exact if $\omega =\ast df$. The subspace ${\Gamma }_{e0}$ captures only those exact differentials approximable by compactly-supported potentials, which on an open surface is a strict subspace of all exact differentials; this gap is the whole content of the open-surface theory and is absent in the compact case 06.04.03.

Non-example. On the plane $\mathbb{C}$, the harmonic differential $dx$ is exact, $dx=d(x)$, yet $x$ is not compactly supported and $dx\notin \Gamma$ at all (infinite energy). The tapered version from the Beginner worked example is in $\Gamma$ but is no longer harmonic. So "exact" and "lies in ${\Gamma }_{e0}$" diverge precisely because compact support is demanded.

Key theorem with proof Intermediate+

Theorem (orthogonal decomposition of $\Gamma$; Weyl 1940, Ahlfors-Sario). Let $X$ be a Riemann surface. Then the Hilbert space of square-integrable differentials decomposes as an orthogonal direct sum of three closed subspaces, $$ \Gamma ;=; \Gamma_h ;\oplus; \Gamma_{e0} ;\oplus; \Gamma_{e0}^, $$ where ${\Gamma }_h$ is the space of harmonic differentials, ${\Gamma }_{e0}$ the closure of differentials of compactly-supported functions, and $\Gamma_{e0}^ = \Gamma_{e0}$.Thestar$$preserves$\Gamma_h$andinterchanges$\Gamma_{e0}$with$\Gamma_{e0}^$.*

Proof. The argument is in four steps: record the star isometry, compute the orthogonal complement of ${\Gamma }_{e0}$ by integration by parts, invoke Weyl's lemma to identify the resulting weak-harmonic space with genuine harmonic differentials, and split off the co-exact summand.

Step 1 — the star is an isometry. From $\ast \ast =-1$ on $1$-forms, $\omega \wedge \overline{\ast \theta }=(\ast \omega )\wedge \overline{\ast (\ast \theta )}$, so $(\ast \omega ,\ast \theta )=(\omega ,\theta )$ and $\ast$ is an orthogonal involution-up-to-sign on $\Gamma$. Hence ${\Gamma }_{e0}^{\ast }=\ast {\Gamma }_{e0}$ is a closed subspace isometric to ${\Gamma }_{e0}$, and any orthogonality between ${\Gamma }_{e0}$ and a star-invariant space transfers to ${\Gamma }_{e0}^{\ast }$.

Step 2 — the complement of ${\Gamma }_{e0}$. A differential $\omega \in \Gamma$ is orthogonal to ${\Gamma }_{e0}$ iff $(\omega ,df)=0$ for every $f\in C_c^{\infty }(X)$. Expanding with $\ast \ast =-1$, $$ (\omega, df) = \iint_X \omega \wedge \overline{,df} = -\iint_X {}\omega \wedge \overline{df}. $$ Integration by parts against the compactly-supported $f$ — no boundary term, since $f$ vanishes outside a compact set — turns the right side into ${\iint }_{X}f\overline{d\ast \omega }$ in the distributional sense. This is zero for all such $f$ exactly when $d\ast \omega =0$ weakly, that is, $\omega$ is weakly co-closed. Therefore $$ \Gamma_{e0}^{\perp} = { \omega \in \Gamma : d{*}\omega = 0 \text{ weakly} }. $$

Step 3 — Weyl's lemma. Apply Step 2 to the star image. Since $\ast$ is an isometry, $\omega \perp {\Gamma }_{e0}^{\ast }$ iff $\ast \omega \perp {\Gamma }_{e0}$ iff $d\ast (\ast \omega )=-d\omega =0$ weakly. So $$ \Gamma_{e0}^\perp \cap (\Gamma_{e0}^)^\perp = { \omega \in \Gamma : d\omega = 0 \text{ and } d{}\omega = 0 \text{ weakly} }. $$ A differential weakly closed and weakly co-closed is, in a local conformal coordinate, a pair of $L^2$ functions satisfying the Cauchy-Riemann system in the distributional sense; Weyl's lemma states that such a distributional solution is smooth, hence a genuine harmonic differential. This is the regularity input that promotes the weak space to ${\Gamma }_h$ and makes ${\Gamma }_h$ the actual harmonic differentials rather than a merely formal kernel.

Step 4 — assembling the sum. ${\Gamma }_{e0}$ and ${\Gamma }_{e0}^{\ast }$ are orthogonal: for $f,g\in C_c^{\infty }(X)$, $$ (df, dg) = \iint_X df \wedge \overline{({}dg)} = -\iint_X df \wedge \overline{dg} = 0, $$ the last equality because $df\wedge dg$ is exact with compactly-supported primitive and integrates to zero by Stokes. Passing to closures keeps the orthogonality. Thus $\Gamma_{e0} \oplus \Gamma_{e0}^$isaclosedsubspace,anditsorthogonalcomplementis$\Gamma_{e0}^\perp \cap (\Gamma_{e0}^)^\perp = \Gamma_h$byStep3.TheHilbert-spaceprojectiontheoremgives$\Gamma = (\Gamma_{e0} \oplus \Gamma_{e0}^) \oplus \Gamma_h$,whichistheclaimeddecomposition.Thestarpreserves$\Gamma_h$becausetheharmonicityconditions$d\omega = 0$,$d{}\omega = 0$aresymmetricunder$\omega \mapsto \omega$(using${} = -1$),andinterchanges$\Gamma_{e0}$with$\Gamma_{e0}^*$bydefinition.$\square$

The orthogonal-projection organisation follows Ahlfors-Sario Ch. II ^{[Ahlfors-Sario]}, who build the entire existence theory of differentials on open surfaces on this single split; de Rham ^{[de Rham]} reorganises the same content around currents and a regularization operator, reaching the decomposition by smoothing distributional harmonic forms rather than by Weyl's lemma directly.

Bridge. This decomposition builds toward the existence theorems of 06.11.03 and 06.11.04, where a harmonic differential with prescribed periods or singularities is produced by projecting a hand-built model differential onto ${\Gamma }_h$; the foundational reason such a projection lands on a smooth harmonic object is exactly Weyl's lemma, supplied here. The split appears again in the classification of open surfaces into parabolic and hyperbolic type, governed by whether certain harmonic functions of finite energy exist. This is exactly the analytic engine of the compact theory 06.04.03 run on a non-compact base: there the Hodge decomposition $\Gamma ={\Gamma }_h\oplus {\Gamma }_e\oplus {\Gamma }_e^{\ast }$ closes up because compactness forces ${\Gamma }_{e0}={\Gamma }_e$, whereas here the gap between ${\Gamma }_{e0}$ and the full exact space generalises the compact statement and is dual to the failure of every harmonic function to be constant. The bridge is the conformal invariance of $\ast$, which is what lets the same Hilbert-space machine run without a global metric, and putting these together gives the central insight that on an open surface the analysis of harmonic differentials is governed by one orthogonal projection whose smoothness is guaranteed by elliptic regularity.

Exercises Intermediate+

Lean formalization Intermediate+

Mathlib does not currently formalise the real Hilbert space $\Gamma$ of $L^2$ differentials on a (possibly non-compact) Riemann surface, the conformal star $\ast$ on $1$-forms, the subspaces ${\Gamma }_h,{\Gamma }_{e0},{\Gamma }_{e0}^{\ast }$, or Weyl's lemma. A proposed signature, in Lean 4 / Mathlib syntax, sketching the target statement:

-- Sketch only; no current Mathlib coverage. See lean_mathlib_gap.
import Mathlib.Analysis.InnerProductSpace.Projection
import Mathlib.Geometry.Manifold.ComplexManifold

namespace Codex.RiemannSurfaces.OpenSurfaces

variable {X : Type*} [RiemannSurface X]   -- not assumed compact

-- The real Hilbert space of L^2 differentials with the Dirichlet inner product
-- (ω, θ) = ∬_X ω ∧ conj(⋆θ), ⋆ the conformal Hodge star.
noncomputable def L2Differentials : Type _ := sorry
noncomputable instance : InnerProductSpace ℝ (L2Differentials (X := X)) := sorry

-- Harmonic = weakly closed and weakly co-closed (smooth after Weyl's lemma).
noncomputable def Γh  : Submodule ℝ (L2Differentials (X := X)) := sorry
-- Closure of differentials of compactly-supported functions, and its ⋆-image.
noncomputable def Γe0  : Submodule ℝ (L2Differentials (X := X)) := sorry
noncomputable def Γe0' : Submodule ℝ (L2Differentials (X := X)) := sorry  -- ⋆ Γe0

-- Orthogonal decomposition Γ = Γh ⊕ Γe0 ⊕ Γe0'.
theorem orthogonal_decomposition :
    (⊤ : Submodule ℝ (L2Differentials (X := X)))
      = Γh ⊔ Γe0 ⊔ Γe0'
    ∧ Pairwise (fun A B => A ⟂ B) ![Γh, Γe0, Γe0'] := by
  sorry

-- Weyl's lemma: a weakly-harmonic L^2 differential is smooth-harmonic.
theorem weyl_lemma (ω : L2Differentials (X := X))
    (h : WeaklyClosed ω ∧ WeaklyCoClosed ω) : Smooth ω ∧ Harmonic ω := by
  sorry

end Codex.RiemannSurfaces.OpenSurfaces

The proof depends on names absent from Mathlib: the conformal invariance of the star on $1$-forms in two real dimensions, the weak (distributional) exterior derivative on $L^2$ differentials, the orthogonal-complement computation via integration by parts against compactly-supported functions, and Weyl's elliptic-regularity lemma. Each is a candidate Mathlib contribution; until then the unit ships with lean_status: none.

Advanced results Master

The orthogonal decomposition $\Gamma ={\Gamma }_h\oplus {\Gamma }_{e0}\oplus {\Gamma }_{e0}^{\ast }$ is the structural floor of the Ahlfors-Sario theory of differentials on open surfaces, and its refinements organise the entire existence theory. The single subspace ${\Gamma }_{e0}$, the closure of compactly-supported gradients, carries the whole difference between the compact and non-compact cases.

Exact differentials and the second exact subspace. Alongside ${\Gamma }_{e0}$, define ${\Gamma }_e$ as the closure of all smooth exact square-integrable differentials $df$ with $f$ smooth of finite energy but not necessarily compactly supported, and ${\Gamma }_h^e={\Gamma }_h\cap {\Gamma }_e$, the harmonic exact differentials — harmonic differentials that are also $L^2$-limits of full exact differentials. On a compact surface ${\Gamma }_{e0}={\Gamma }_e$ and ${\Gamma }_h^e=0$. On an open surface the chain ${\Gamma }_{e0}\subseteq {\Gamma }_e\subseteq {\Gamma }_{e0}\oplus {\Gamma }_h^e$ holds, and ${\Gamma }_h^e$ measures the harmonic differentials that arise as energy-limits of exact ones. Ahlfors-Sario refine the three-term split into a five-term decomposition $\Gamma ={\Gamma }_{h0}\oplus {\Gamma }_{h0}^{\ast }\oplus {\Gamma }_{he}\oplus {\Gamma }_{e0}\oplus {\Gamma }_{e0}^{\ast }$, separating the harmonic space ${\Gamma }_h$ into the square-integrable harmonic-with-singularity part ${\Gamma }_{he}$ and a residual ${\Gamma }_{h0}\oplus {\Gamma }_{h0}^{\ast }$ tied to the ideal boundary; this finer split is the apparatus that produces differentials with prescribed periods and singularities and is the analytic content of 06.11.03.

Classification of open surfaces by null classes. The decomposition supports the partition of open Riemann surfaces into null classes $O_G,O_{HB},O_{HD},O_{AD},\dots$ — surfaces admitting no non-constant Green's function, no non-constant bounded harmonic function, no non-constant Dirichlet-finite harmonic function, and so on. A surface is parabolic ($O_G$) exactly when the constants are the only bounded harmonic functions of finite energy, equivalently when ${\Gamma }_{h0}=0$ so the harmonic space carries no ideal-boundary contribution. The plane $\mathbb{C}$ and the punctured plane ${\mathbb{C}}^{\ast }$ are parabolic; the unit disc is hyperbolic. The decomposition translates each topological-or-potential-theoretic null condition into the vanishing of a specific summand, making the type problem a question about the geometry of $\Gamma$.

Reproducing differentials and the bilinear relations. On a finite open surface — a compact bordered surface — the harmonic space ${\Gamma }_h$ is finite-dimensional and carries reproducing differentials dual to the homology classes of the boundary and handles. The bilinear relation $(\omega ,\ast \theta )={\iint }_X\omega \wedge \theta$ pairs the harmonic space against itself through the star, and the resulting period matrix obeys reciprocity laws extending Riemann's bilinear relations from the compact case 06.04.03 to surfaces with boundary; the imaginary-part positivity becomes the positive-definiteness of the Dirichlet form on ${\Gamma }_h$.

Weyl's lemma and elliptic regularity. The smoothness step is a special case of interior elliptic regularity: the Laplace-Beltrami operator $\Delta =-(d\ast {)}^2$-type combination is uniformly elliptic in any conformal coordinate, and a distribution annihilated by it lies in every local Sobolev space, hence in $C^{\infty }$ by Sobolev embedding. Weyl's 1940 hypoellipticity argument for the Laplacian — by mollification and a mean-value characterisation — predates and motivates the general hypoellipticity theory of Hörmander; on a surface the two-dimensionality makes the Cauchy-Riemann factorisation available, so weak harmonicity is weak holomorphy of $\partial$-components and Weyl's lemma reduces to the regularity of weakly holomorphic functions.

Synthesis. The orthogonal decomposition is the foundational reason the existence theory on open surfaces reduces to linear algebra in a Hilbert space: every differential with prescribed periods or singularities is reached by orthogonal projection onto ${\Gamma }_h$, and the central insight is that this projection is smooth because of Weyl's lemma. The three-term split generalises the two-term Hodge decomposition of the compact case 06.04.03, to which it collapses when compact support becomes automatic, and the gap between ${\Gamma }_{e0}$ and the full exact space is dual to the existence of non-constant Dirichlet-finite harmonic functions, the dichotomy that classifies surfaces into parabolic and hyperbolic type. Putting these together, the analytic engine of 06.04.05 — harmonic projection in an $L^2$ space governed by an elliptic operator — is exactly this construction read with the metric $\bar{\partial }$-Laplacian in place of the conformal star, so the two unify: this is the same orthogonal-projection method, run once with $d,d\ast$ on real differentials and once with $\bar{\partial },{\bar{\partial }}^{\ast }$ on line-bundle-valued forms. The bridge is conformal invariance, which removes any dependence on a chosen metric and makes the decomposition intrinsic to the complex structure, and the whole apparatus builds toward the uniformization and classification theory of open surfaces, where the null classes appear again as the organising invariants.

Full proof set Master

Proposition (orthogonality of ${\Gamma }_{e0}$ and ${\Gamma }_{e0}^{\ast }$). For any Riemann surface $X$, $\Gamma_{e0} \perp \Gamma_{e0}^$in$\Gamma$.*

Proof. By continuity of the inner product it suffices to check generators. Let $f,g\in C_c^{\infty }(X)$. Then $$ (df, {}dg) = \iint_X df \wedge \overline{({}dg)} = \iint_X df \wedge \overline{(-dg)} = -\iint_X df \wedge \overline{dg}. $$ The $2$-form $df\wedge dg=d(fdg)$ is exact with the compactly-supported primitive $fdg$, so Stokes' theorem on $X$ (no boundary term, by compact support) gives ${\iint }_{X}d(fdg)=0$, and the conjugate integral vanishes identically. Hence $(df, {}dg) = 0$ongenerators,andpassingto$L^2$-closurespreservesorthogonality.$\square$

Proposition (complement of ${\Gamma }_{e0}$). $\Gamma_{e0}^\perp = {\omega \in \Gamma : d{}\omega = 0 \text{ weakly}}$.*

Proof. $\omega \perp {\Gamma }_{e0}$ iff $(\omega ,df)=0$ for all $f\in C_c^{\infty }(X)$, since such $df$ span a dense subset of ${\Gamma }_{e0}$. Compute $$ (\omega, df) = \iint_X \omega \wedge \overline{,df} = -\iint_X ({}\omega) \wedge \overline{df} = \iint_X f, \overline{d{}\omega}, $$ where the middle equality uses $\eta \wedge \overline{,df} = -({}\eta) \wedge \overline{df}$from${} = -1$,andthelastisintegrationbypartswithvanishingboundarytermbecause$f$iscompactlysupported.Thepairing$\iint_X f, \overline{d{}\omega}$vanishesforeverytestfunction$f$iffthedistribution$d{*}\omega$iszero,i.e.$\omega$isweaklyco-closed.$\square$

Proposition (Weyl's lemma identifies the harmonic summand). The space $W := {\omega \in \Gamma : d\omega = 0 \text{ and } d{}\omega = 0 \text{ weakly}}$consistsofsmoothharmonicdifferentials,so$W = \Gamma_h$.*

Proof. In a conformal coordinate $z=x+iy$ write $\omega =adx+bdy$ with $a,b\in L_{\mathrm{loc}}^2$. The conditions $d\omega =0$ and $d\ast \omega =0$ read, distributionally, ${\partial }_{x}b-{\partial }_{y}a=0$ and ${\partial }_{x}a+{\partial }_{y}b=0$ — the Cauchy-Riemann system for the pair $(a,-b)$, equivalently weak holomorphy of $a-ib$ as a function of $z$. Weyl's lemma for the Laplacian (or, here, for $\bar{\partial }$): a locally integrable function weakly annihilated by the Laplacian agrees almost everywhere with a smooth harmonic function. Applying it to the real and imaginary parts shows $a,b$ are smooth, so $\omega$ is a smooth $1$-form satisfying $d\omega =d\ast \omega =0$ classically, hence harmonic. Conversely every smooth harmonic differential lies in $W$. Therefore $W={\Gamma }_h$. $\square$

Theorem (orthogonal decomposition). $\Gamma = \Gamma_h \oplus \Gamma_{e0} \oplus \Gamma_{e0}^$orthogonally,with$$preserving$\Gamma_h$andinterchanging$\Gamma_{e0}$and$\Gamma_{e0}^$.*

Proof. By the first proposition ${\Gamma }_{e0}\perp {\Gamma }_{e0}^{\ast }$, so $S:={\Gamma }_{e0}\oplus {\Gamma }_{e0}^{\ast }$ is a closed subspace and $\Gamma =S\oplus S^{\perp }$ by the Hilbert-space projection theorem. Now $S^{\perp }={\Gamma }_{e0}^{\perp }\cap ({\Gamma }_{e0}^{\ast }{)}^{\perp }$. The second proposition gives ${\Gamma }_{e0}^{\perp }=\{d\ast \omega =0\}$; applying it to $\ast \omega$ and using that $\ast$ is an isometry with $\ast \ast =-1$ gives $({\Gamma }_{e0}^{\ast }{)}^{\perp }=\{\omega :d\omega =0\text{ weakly}\}$. Hence $S^{\perp }=\{\omega :d\omega =0,d\ast \omega =0\text{ weakly}\}=W$, which the third proposition identifies with ${\Gamma }_h$. The star preserves ${\Gamma }_h$ because the defining conditions are symmetric under $\omega \mapsto \ast \omega$, and interchanges the two exact summands by construction. $\square$

Corollary (Hodge decomposition as the compact specialisation). If $X$ is compact, $\Gamma = \Gamma_h \oplus \Gamma_e \oplus \Gamma_e^$with$\Gamma_e$theclosureofallexact$L^2$ differentials, recovering 06.04.03.*

Proof. On a compact $X$ every smooth function is compactly supported, so ${\Gamma }_{e0}={\Gamma }_e$ and ${\Gamma }_{e0}^{\ast }={\Gamma }_e^{\ast }$. Substituting into the theorem yields the classical two-exact-summand Hodge orthogonal decomposition of $L^2$ $1$-forms, whose harmonic summand is the finite-dimensional space of harmonic differentials computed in 06.04.03. $\square$

Corollary (harmonic projection is energy-minimising). For closed $\sigma \in \Gamma$, the ${\Gamma }_h$-component ${\sigma }_h$ is the unique minimiser of $\Vert \sigma -\eta \Vert$ over $\eta \in {\Gamma }_{e0}$.

Proof. Closedness gives $\sigma \in {\Gamma }_h\oplus {\Gamma }_{e0}$ (it is orthogonal to ${\Gamma }_{e0}^{\ast }$ by the second proposition applied to $\ast \sigma$). Write $\sigma ={\sigma }_h+{\sigma }_{e0}$. For $\eta \in {\Gamma }_{e0}$, $\Vert \sigma -\eta {\Vert }^2=\Vert {\sigma }_h{\Vert }^2+\Vert {\sigma }_{e0}-\eta {\Vert }^2\ge \Vert {\sigma }_h{\Vert }^2$, with equality iff $\eta ={\sigma }_{e0}$. Thus ${\sigma }_h=\sigma -{\sigma }_{e0}$ is the foot of the perpendicular from $\sigma$ to ${\Gamma }_{e0}$, the unique energy-minimiser. $\square$

Connections Master

• Square-integrable differentials and the conformal star 06.11.01. The present unit takes the $L^2$ differentials and the conformal Hodge star constructed in 06.11.01 as its objects and shows that the resulting Hilbert space carries a canonical three-term orthogonal decomposition. The metric-independence of $\ast$ on $1$-forms established there is what makes the inner product $(\omega ,\theta )={\iint }_X\omega \wedge \overline{\ast \theta }$ intrinsic, so the decomposition is an invariant of the complex structure rather than of a chosen metric.

• Hodge decomposition on a compact Riemann surface 06.04.03. The compact Hodge decomposition $\Gamma ={\Gamma }_h\oplus {\Gamma }_e\oplus {\Gamma }_e^{\ast }$ is the specialisation of the open-surface split when compact support becomes automatic. The open-surface theory isolates the subspace ${\Gamma }_{e0}$ of compactly-supported gradients, whose strict containment in the full exact space vanishes on a compact base; the present unit is the non-compact generalisation, and 06.04.03 is the case to which it collapses.

• Hilbert-space PDE for $\bar{\partial }$ 06.04.05. Both units run Weyl's orthogonal-projection method in an $L^2$ space governed by an elliptic operator. There the operator is the $\bar{\partial }$-Laplacian on line-bundle-valued forms and the harmonic kernel computes sheaf cohomology; here it is the real Laplacian written through the conformal star and the harmonic space computes period and boundary data. The two are the holomorphic and the real-harmonic faces of one analytic engine, joined by the Cauchy-Riemann factorisation in real dimension two.

• Existence of harmonic differentials with prescribed periods 06.11.03. The refined Ahlfors-Sario decomposition built on the split of this unit is the apparatus that produces a harmonic differential realising a prescribed period homomorphism by projecting a compactly-supported model onto ${\Gamma }_h$; Weyl's lemma supplies the smoothness of the result.

• Classification of open surfaces by null classes 06.11.04. The partition of open surfaces into parabolic and hyperbolic type — and the finer null classes $O_G,O_{HB},O_{HD}$ — is read off the geometry of $\Gamma$: each null condition is the vanishing of a specific summand of the harmonic space, so the type problem becomes a statement about the orthogonal decomposition constructed here.

Historical & philosophical context Master

Hermann Weyl introduced the orthogonal-projection method in his 1940 paper The method of orthogonal projection in potential theory ^{[Weyl 1940]} (Comment. Math. Helv. 7, 411-444), recasting the Dirichlet principle as an orthogonal projection in a Hilbert space of square-integrable vector fields rather than as a direct minimisation of the energy integral. The reformulation removed the logical gap in Riemann's original Dirichlet-principle argument — the assumption that the energy infimum is attained — by replacing it with the completeness of the $L^2$ space and the elementary existence of orthogonal projections; the harmonic field sought is the projection of a test field onto the closed subspace of divergence-and-curl-free fields. Weyl's smoothness step, that a weakly harmonic field is genuinely smooth, is the result now called Weyl's lemma and is the prototype of elliptic hypoellipticity; he returned to the regularity question in his 1943 On Hodge's theory of harmonic integrals ^{[Weyl 1943]} (Ann. of Math. (2) 44, 1-6) while repairing the analytic gap in Hodge's compact-manifold theory.

Lars Ahlfors and Leo Sario systematised the open-surface theory in their 1960 monograph Riemann Surfaces ^{[Ahlfors-Sario]}, taking the orthogonal decomposition of $L^2$ differentials as the structural foundation and refining it into the five-term split that produces differentials with prescribed periods and singularities and that classifies open surfaces by null classes. Georges de Rham's Variétés différentiables ^{[de Rham]} developed the parallel theory of currents and a regularization operator, reaching the Hodge-de Rham decomposition by smoothing distributional harmonic forms; Halsey Royden's later treatment ^[Royden] of the $L^2$ differentials emphasised the role of the compactly-supported exact subspace ${\Gamma }_{e0}$ as the precise carrier of the difference between the compact and open cases. Springer's 1957 textbook gave the first widely-used graduate exposition of the Hilbert-space-of-differentials approach for the curve case.

Bibliography Master

@article{Weyl1940OrthogonalProjection,
  author  = {Weyl, Hermann},
  title   = {The method of orthogonal projection in potential theory},
  journal = {Comment. Math. Helv.},
  volume  = {7},
  year    = {1940},
  pages   = {411--444}
}

@article{Weyl1943HodgeIntegrals,
  author  = {Weyl, Hermann},
  title   = {On {Hodge's} theory of harmonic integrals},
  journal = {Ann. of Math. (2)},
  volume  = {44},
  year    = {1943},
  pages   = {1--6}
}

@book{AhlforsSario1960,
  author    = {Ahlfors, Lars V. and Sario, Leo},
  title     = {Riemann Surfaces},
  series    = {Princeton Mathematical Series},
  volume    = {26},
  publisher = {Princeton University Press},
  year      = {1960}
}

@book{deRham1955,
  author    = {de Rham, Georges},
  title     = {Vari{\'e}t{\'e}s diff{\'e}rentiables: formes, courants, formes harmoniques},
  publisher = {Hermann},
  year      = {1955}
}

@book{Springer1957,
  author    = {Springer, George},
  title     = {Introduction to Riemann Surfaces},
  publisher = {Addison-Wesley},
  year      = {1957}
}

@book{Royden1956FunctionTheory,
  author    = {Royden, Halsey L.},
  title     = {Function theory on compact Riemann surfaces and the $L^2$ theory of differentials},
  publisher = {note: collected lectures / later treatment},
  year      = {1956}
}

@book{Forster1981,
  author    = {Forster, Otto},
  title     = {Lectures on Riemann Surfaces},
  series    = {Graduate Texts in Mathematics},
  volume    = {81},
  publisher = {Springer},
  year      = {1981}
}