[proofplan] By the [Bochner Vanishing Theorem](/theorems/2760) Part (3), the hypotheses $\operatorname{Ric}(g) \geq 0$ and $b^1(M) = n$ produce $n$ parallel $1$-forms $\alpha_1, \ldots, \alpha_n$ that span $T^*M$ at every point and a flat metric on $M$. We construct an explicit isometry from $M$ to a flat torus by integrating these $1$-forms. Since each $\alpha_i$ is closed, the line integral $\int_p^x \alpha_i$ depends only on the homotopy class of the path from $p$ to $x$, and the periods over closed loops generate a discrete lattice $\Lambda \subset \mathbb{R}^n$ via the homomorphism $H_1(M; \mathbb{Z}) \to \mathbb{R}^n$ produced by the period map. Define $F: M \to \mathbb{R}^n / \Lambda$ by $F(x) = (\int_p^x \alpha_i)_i \pmod \Lambda$. Because the $\alpha_i$ form a parallel coframe, $F$ has invertible differential and is a local isometry between flat metrics; on compact $M$, $F$ is a finite covering map; the universal cover of any flat torus is a flat torus, forcing $M$ itself to be a flat torus. [/proofplan] [step:Extract a parallel coframe and a flat metric from the Bochner hypotheses] Let $(M, g)$ be a compact connected oriented Riemannian manifold of dimension $n$ with $\operatorname{Ric}(g) \geq 0$ and $b^1(M) = n$. By the [Bochner Vanishing Theorem](/theorems/2760), Parts (2) and (3): \begin{enumerate} \item Every harmonic $1$-form $\alpha \in \mathcal{H}^1(M)$ is parallel ($\nabla\alpha = 0$). \item There exists a basis $\alpha_1, \ldots, \alpha_n \in \mathcal{H}^1(M)$, each parallel. \item The metric $g$ is flat: the Riemann curvature tensor $R$ vanishes identically on $M$. \end{enumerate} The values $(\alpha_i)_x$ form a basis of $T_x^* M$ for every $x \in M$ (parallel transport is a linear isomorphism on each cotangent fibre, and the values $(\alpha_i)_{x_0}$ at a single point form a basis by the saturation of $b^1(M) \leq n$ in Part (2) of the cited theorem). So $\{\alpha_1, \ldots, \alpha_n\}$ is a global parallel coframe on $T^*M$. Each $\alpha_i$ is a harmonic $1$-form on the closed $M$, hence closed: by the [Harmonic Iff Closed and Co-closed](/theorems/2744) characterisation, $d\alpha_i = 0$ and $\delta\alpha_i = 0$ for every $i$. We will use closedness for path-independence and the parallel coframe property for local isometry below. To make the parallel coframe orthonormal with respect to $g$, apply the Gram–Schmidt process to $(\alpha_1)_{x_0}, \ldots, (\alpha_n)_{x_0}$ in the inner product space $T_{x_0}^* M$ (with the cometric inherited from $g$). Each Gram–Schmidt step is a linear combination of the original $\alpha_i$'s with constant coefficients, which preserves both the parallel and harmonic properties (constant linear combinations commute with $\nabla$ and $\Delta$). After this step, we claim $\langle \alpha_i, \alpha_j\rangle_g \equiv \delta_{ij}$ on $M$. Indeed, since the Levi-Civita connection is metric-compatible, the cometric on $T^*M$ inherits the Leibniz rule: for any $X \in \mathfrak{X}(M)$, \begin{align*} X\langle \alpha_i, \alpha_j\rangle = \langle \nabla_X \alpha_i, \alpha_j\rangle + \langle \alpha_i, \nabla_X \alpha_j\rangle. \end{align*} Both $\nabla_X \alpha_i = 0$ and $\nabla_X \alpha_j = 0$ since the $\alpha_i$ are parallel, so $X\langle \alpha_i, \alpha_j\rangle \equiv 0$ for every vector field $X$. Hence $\langle \alpha_i, \alpha_j\rangle_g$ is constant on the connected manifold $M$, and the constant equals $\delta_{ij}$ because Gram–Schmidt orthonormalises at $x_0$. [/step] [step:Construct the period lattice via the de Rham period map] Fix $p \in M$. For each $i$ and each piecewise-smooth path $\gamma: [0, 1] \to M$ with $\gamma(0) = p$, the line integral $\int_\gamma \alpha_i \in \mathbb{R}$ is defined by \begin{align*} \int_\gamma \alpha_i := \int_0^1 \alpha_i(\dot\gamma(t))\,d\mathcal{L}^1(t). \end{align*} Because each $\alpha_i$ is closed, the integral $\int_\gamma \alpha_i$ depends only on the relative homotopy class of $\gamma$ rel endpoints: by Stokes' theorem applied to a smooth homotopy $H: [0, 1]^2 \to M$ between two paths $\gamma_0$ and $\gamma_1$ (both starting at $p$, ending at the same point), \begin{align*} \int_{\gamma_1} \alpha_i - \int_{\gamma_0} \alpha_i = \int_{\partial H([0,1]^2)} \alpha_i = \int_{H([0,1]^2)} d\alpha_i = 0. \end{align*} Define the **period homomorphism** \begin{align*} \Pi: \pi_1(M, p) &\to \mathbb{R}^n, \\ [\gamma] &\mapsto \Big(\int_\gamma \alpha_1, \ldots, \int_\gamma \alpha_n\Big), \end{align*} which factors through $H_1(M; \mathbb{Z})$ since $\mathbb{R}^n$ is abelian: $\Pi$ vanishes on the commutator subgroup of $\pi_1(M, p)$, so descends to a homomorphism $\bar\Pi: H_1(M; \mathbb{Z}) \to \mathbb{R}^n$. Set $\Lambda := \bar\Pi(H_1(M; \mathbb{Z})) \subseteq \mathbb{R}^n$. We claim $\Lambda$ is a discrete subgroup of rank $n$ — that is, a full-rank lattice. **Factoring through the free part.** Since $M$ is compact, $H_1(M; \mathbb{Z})$ is finitely generated. Because $\mathbb{R}^n$ is torsion-free, the homomorphism $\bar\Pi: H_1(M; \mathbb{Z}) \to \mathbb{R}^n$ kills the torsion subgroup $T \subseteq H_1(M; \mathbb{Z})$ and descends to a $\mathbb{Z}$-linear map \begin{align*} \Phi: H_1(M; \mathbb{Z}) / T \to \mathbb{R}^n \end{align*} with the same image $\Lambda$. The free abelian group $H_1(M; \mathbb{Z}) / T$ has rank $b_1(M) = n$ (here we use Poincaré duality / the universal coefficient theorem: $b_1(M) = \dim H^1_{\mathrm{dR}}(M; \mathbb{R}) = n$ by hypothesis). Choosing a $\mathbb{Z}$-basis identifies $H_1(M; \mathbb{Z}) / T \cong \mathbb{Z}^n$, so $\Phi: \mathbb{Z}^n \to \mathbb{R}^n$ is a $\mathbb{Z}$-linear map whose image is $\Lambda$. **Injectivity of $\Phi$ and full $\mathbb{R}$-span.** We show $\Phi$ is injective by showing the $\mathbb{R}$-linear extension \begin{align*} \Phi_\mathbb{R}: H_1(M; \mathbb{R}) &\to \mathbb{R}^n, \\ [\gamma] &\mapsto \Big(\int_\gamma \alpha_1, \ldots, \int_\gamma \alpha_n\Big), \end{align*} is an isomorphism. By the [Hodge Decomposition](/theorems/2745), the harmonic representatives $\alpha_1, \ldots, \alpha_n$ are linearly independent in $H^1_{\mathrm{dR}}(M)$ (they are pointwise linearly independent — parallel forms with linearly independent values at one point have linearly independent values everywhere — hence in particular linearly independent as cohomology classes); since $\dim H^1_{\mathrm{dR}}(M) = b^1(M) = n$, they form a basis of $H^1_{\mathrm{dR}}(M)$. By the de Rham theorem, the pairing \begin{align*} H^1_{\mathrm{dR}}(M) \times H_1(M; \mathbb{R}) \to \mathbb{R}, \qquad ([\alpha], [\gamma]) \mapsto \int_\gamma \alpha, \end{align*} is non-degenerate, so the period map $\Phi_\mathbb{R}$ taking $[\gamma] \mapsto (\int_\gamma \alpha_i)_{i=1}^n$ is a linear isomorphism (it is the matrix of the pairing in dual bases). Since $\Phi = \Phi_\mathbb{R}|_{\mathbb{Z}^n}$ via the inclusion $\mathbb{Z}^n \hookrightarrow H_1(M; \mathbb{R})$ (free abelian group inside its real-vector-space tensor product), $\Phi$ is injective and $\Lambda = \Phi(\mathbb{Z}^n)$ has $\mathbb{R}$-span equal to $\mathbb{R}^n$. **Discreteness as a $\mathbb{Z}$-lattice.** Since $\Phi: \mathbb{Z}^n \to \mathbb{R}^n$ is an injective $\mathbb{Z}$-linear map whose image $\mathbb{R}$-spans $\mathbb{R}^n$, the vectors $\Phi(e_1), \ldots, \Phi(e_n)$ are $\mathbb{R}$-linearly independent in $\mathbb{R}^n$ (any nontrivial $\mathbb{R}$-linear relation would contradict the $\mathbb{R}$-span being $n$-dimensional combined with there being only $n$ generators). Hence $\Phi(e_1), \ldots, \Phi(e_n)$ form an $\mathbb{R}$-basis of $\mathbb{R}^n$, and \begin{align*} \Lambda = \Phi(\mathbb{Z}^n) = \mathbb{Z}\Phi(e_1) \oplus \cdots \oplus \mathbb{Z}\Phi(e_n) \end{align*} is the $\mathbb{Z}$-span of an $\mathbb{R}$-basis. This is the standard definition of a full-rank lattice in $\mathbb{R}^n$, and such subgroups are automatically discrete (the $\mathbb{Z}$-span of an $\mathbb{R}$-basis $v_1, \ldots, v_n$ is discrete because the map $\mathbb{R}^n \to \mathbb{R}^n$ sending $e_i \mapsto v_i$ is a linear isomorphism carrying the standard discrete lattice $\mathbb{Z}^n$ to $\Lambda$). The quotient $T^n := \mathbb{R}^n / \Lambda$ inherits the standard flat metric from $\mathbb{R}^n$, and is an oriented closed Riemannian $n$-manifold isometric to a flat $n$-torus. [/step] [step:Define the global map $F: M \to \mathbb{R}^n / \Lambda$ and show it is well-defined] For $x \in M$, choose any piecewise-smooth path $\gamma_x: [0, 1] \to M$ with $\gamma_x(0) = p$, $\gamma_x(1) = x$. Define \begin{align*} F: M &\to \mathbb{R}^n / \Lambda, \\ x &\mapsto \Big(\int_{\gamma_x} \alpha_1, \ldots, \int_{\gamma_x} \alpha_n\Big) \pmod \Lambda. \end{align*} The value $F(x)$ is independent of the choice of $\gamma_x$: if $\gamma_x'$ is another path from $p$ to $x$, the loop $\gamma_x \cdot (\gamma_x')^{-1}$ is an element $[\delta] \in \pi_1(M, p)$, and \begin{align*} \int_{\gamma_x} \alpha_i - \int_{\gamma_x'} \alpha_i = \int_{\delta} \alpha_i = (\Pi[\delta])_i \in \Lambda_i, \end{align*} which is in the lattice $\Lambda$ component by component. So the difference projects to $0 \in \mathbb{R}^n / \Lambda$, and $F(x)$ is unambiguously defined. $F$ is smooth: pick a local lift $\tilde F: U \to \mathbb{R}^n$ on a simply connected neighbourhood $U \ni x$, computed via paths inside $U$; smooth integrals of smooth $1$-forms yield a smooth map. The projection $\mathbb{R}^n \to \mathbb{R}^n / \Lambda$ is a local diffeomorphism, so $F = \pi \circ \tilde F$ is smooth. [/step] [step:Show $F$ is a local isometry] We compute the differential $dF_x: T_x M \to T_{F(x)}(\mathbb{R}^n / \Lambda) \cong \mathbb{R}^n$ in the local lift $\tilde F$. For $X \in T_x M$, choose any smooth path $\eta(t)$ in $M$ with $\eta(0) = x$, $\dot\eta(0) = X$; then by the fundamental theorem of calculus applied to $t \mapsto \int_p^{\eta(t)} \alpha_i$, \begin{align*} (d\tilde F_x(X))_i = \frac{d}{dt}\bigg|_{t=0} \int_p^{\eta(t)} \alpha_i = \alpha_i(\dot\eta(0)) = \alpha_i(X). \end{align*} So $dF_x(X) = (\alpha_1(X), \ldots, \alpha_n(X)) \in \mathbb{R}^n$ under the local lift identification. Since $\{\alpha_1, \ldots, \alpha_n\}$ is a parallel **orthonormal** coframe at every point of $M$ (after the Gram-Schmidt step, $\langle \alpha_i, \alpha_j\rangle_g \equiv \delta_{ij}$), the dual frame $\{X_1, \ldots, X_n\}$ is a global parallel orthonormal frame on $TM$. The differential $dF_x$ sends $X_i|_x \mapsto e_i \in \mathbb{R}^n$ (the standard basis), since $\alpha_j(X_i) = \delta_{ij}$. This $dF_x$ is a linear isometry $T_x M \to \mathbb{R}^n$ between Euclidean inner product spaces. The metric on $\mathbb{R}^n / \Lambda$ is the standard Euclidean metric (descending from $\mathbb{R}^n$), so $dF_x$ is also a linear isometry $T_x M \to T_{F(x)}(\mathbb{R}^n / \Lambda)$. Therefore $F$ is a local isometry between the flat Riemannian manifolds $(M, g)$ and $(T^n, g_{\mathrm{flat}})$. In particular, $dF_x$ is a linear isomorphism at every $x$, so $F$ is a **local diffeomorphism**. [/step] [step:Conclude that $F$ is a covering map and an isometry] We invoke the [Local Isometry from Complete Manifold is a Covering Map](/theorems/2735) theorem: if $f: (M, g) \to (N, h)$ is a local isometry between connected Riemannian manifolds, $M$ is geodesically complete, and $N$ is connected, then $f$ is a Riemannian covering map. Verify the hypotheses: \begin{itemize} \item $M$ is connected (by hypothesis) and oriented; $T^n = \mathbb{R}^n / \Lambda$ is connected. \item $F: M \to T^n$ is a local isometry, established in the previous step. \item $M$ is geodesically complete: $M$ is compact, and by the [Hopf–Rinow Theorem](/theorems/2726), every compact Riemannian manifold is geodesically complete. \end{itemize} All hypotheses are satisfied, so $F$ is a **Riemannian covering map** $M \to T^n$. Since $M$ is compact, the cover $F$ has finitely many sheets, say $k = |F^{-1}(F(p))|$. Lifting $F$ to universal covers gives a smooth map $\widetilde{F}: \widetilde{M} \to \widetilde{T^n} = \mathbb{R}^n$ between simply connected Riemannian manifolds. Equip $\widetilde{M}$ with the pulled-back Riemannian metric $\pi^*g$, where $\pi: \widetilde{M} \to M$ is the universal covering: this is a flat metric and $\pi$ is a local isometry. Then $\widetilde{F}$ is a local isometry between connected Riemannian manifolds (composition $F \circ \pi$ on small enough neighbourhoods is a local isometry, as both factors are), and $\widetilde{M}$ is geodesically complete because the local isometry $\pi: (\widetilde{M}, \pi^*g) \to (M, g)$ pulls back complete geodesics from $M$ (compact $\Rightarrow$ complete by [Hopf–Rinow](/theorems/2726)) to complete geodesics on $\widetilde{M}$. By the [Local Isometry from Complete Manifold is a Covering Map](/theorems/2735) theorem applied to $\widetilde{F}$, $\widetilde{F}$ is a Riemannian covering map. Both $\widetilde{M}$ and $\mathbb{R}^n$ are simply connected, so the only connected covering of $\mathbb{R}^n$ from a simply connected space is the identity-up-to-isomorphism: $\widetilde{F}$ is a Riemannian isometry. In particular, $\widetilde{M} \cong \mathbb{R}^n$ as Riemannian manifolds. **Deck transformations act by translations.** The deck transformation group $G \cong \pi_1(M)$ acts on $\widetilde{M}$ by isometries (the action lifts $F$ to $\widetilde{F}$, and $F$-deck-transformations descend by hypothesis to identity-on-$T^n$). Transferring through the isometry $\widetilde{F}$, the same group $G$ acts on $\mathbb{R}^n$ by isometries — that is, by Euclidean motions $x \mapsto Ax + b$ with $A \in \mathrm{O}(n)$, $b \in \mathbb{R}^n$. We claim each $g \in G$ acts as a pure translation, i.e., $A = I$. The pulled-back coframe on $\widetilde{M}$ is $\widetilde{\alpha}_i := \pi^* \alpha_i$, a parallel orthonormal coframe on $\widetilde{M}$. Under the isometry $\widetilde{F}: \widetilde{M} \to \mathbb{R}^n$, this coframe corresponds to the parallel orthonormal coframe pushed forward to $\mathbb{R}^n$, which by the differential computation of step 4 is the **standard** coframe $dy_1, \ldots, dy_n$ on $\mathbb{R}^n$ (since $d\widetilde{F}$ sends the dual frame $\widetilde{X}_i$ to the standard basis $e_i$, by the local computation $dF_x(X) = (\alpha_1(X), \ldots, \alpha_n(X))$ holding for $\widetilde{F}$ as well). The $G$-action on $\widetilde{M}$ preserves $\widetilde{\alpha}_i$ (since $\widetilde{\alpha}_i = \pi^* \alpha_i$ is by construction $\pi$-equivariant: $g^* \widetilde{\alpha}_i = g^* \pi^* \alpha_i = (\pi \circ g)^* \alpha_i = \pi^* \alpha_i = \widetilde{\alpha}_i$). Transferring through $\widetilde{F}$, the action of $G$ on $\mathbb{R}^n$ preserves the standard coframe $dy_1, \ldots, dy_n$. For an isometry $g(y) = Ay + b$ of $\mathbb{R}^n$, the pullback $g^*(dy_i) = \sum_j A_{ji} \, dy_j$. The condition $g^*(dy_i) = dy_i$ for every $i$ forces $A_{ji} = \delta_{ji}$, i.e., $A = I$. So $g(y) = y + b$ is a translation, and $G$ acts by translations on $\mathbb{R}^n$. **Identifying the translation lattice.** The translation subgroup $G \subseteq \mathbb{R}^n$ acts freely and properly discontinuously, hence is a discrete subgroup of $\mathbb{R}^n$. Its quotient is $\mathbb{R}^n / G \cong \widetilde{M}/G = M$, isometrically. Since $M$ is compact, $G$ is a full-rank lattice in $\mathbb{R}^n$ (a discrete subgroup of $\mathbb{R}^n$ has compact quotient $\iff$ it is a full-rank lattice). Setting $\Lambda_M := G$, we conclude $M$ is isometric to the flat torus $\mathbb{R}^n / \Lambda_M$, with $\Lambda_M$ a full-rank lattice in $\mathbb{R}^n$. This completes the proof. [guided] The proof has a clean three-act structure: (1) extract a global parallel orthonormal coframe from the Bochner hypotheses, (2) integrate the coframe to define a map to a torus, (3) prove the map is an isometric covering, hence an isometry from $M$ to a flat torus. **Why the Bochner output is exactly what we need.** [Bochner Vanishing Theorem](/theorems/2760) Parts (2) and (3) handed us: - $n$ linearly independent parallel $1$-forms $\alpha_1, \ldots, \alpha_n$. - A flat metric on $M$. A parallel $1$-form is closed (since $d\alpha = \sum \nabla\alpha = 0$), and parallel forms can always be made orthonormal globally by a single Gram–Schmidt step at one point (parallel transport preserves the inner product). So we get a parallel orthonormal coframe — the maximum amount of structure one could ask for. **Why integration produces a map to a torus.** A closed $1$-form on a compact manifold has a well-defined "period" against each homology class (de Rham). The collection of periods $\Pi: \pi_1(M, p) \to \mathbb{R}^n$ — using all $n$ forms simultaneously — is a homomorphism (because $\mathbb{R}^n$ is abelian and the action of $\pi_1$ on $\mathbb{R}$ via the period factors through the abelianisation $H_1(M; \mathbb{Z})$). Its image $\Lambda$ is a finitely generated discrete subgroup of $\mathbb{R}^n$. The de Rham theorem combined with $\dim H^1_{\mathrm{dR}}(M) = n$ tells us $\Lambda$ has rank $n$ — the period map $H_1(M; \mathbb{R}) \to \mathbb{R}^n$ is an isomorphism, so $\Lambda$ spans $\mathbb{R}^n$ over $\mathbb{R}$, and discreteness then ensures $\Lambda$ is a *full-rank lattice* (rather than a dense subgroup). The map $F: M \to \mathbb{R}^n / \Lambda$ via $F(x) = (\int_p^x \alpha_i)_i \pmod \Lambda$ is well-defined precisely because the periods generate $\Lambda$. **Why $F$ is a local isometry.** The differential of $F$ in a local lift is $dF_x(X) = (\alpha_1(X), \ldots, \alpha_n(X))$, computed by differentiating $\int_p^{\eta(t)} \alpha_i$ at $t = 0$ (FTC for line integrals). Since the $\alpha_i$ form a parallel **orthonormal** coframe, $dF_x$ is a linear isometry from $T_x M$ (with the metric $g$) to $\mathbb{R}^n$ (with the Euclidean metric, which is the metric on $T^n$ via the Euclidean covering). So $F$ is a local isometry. **Why a local isometry from compact to compact is a covering.** The general principle is the Hopf–Rinow theorem combined with the fact that local isometries of complete manifolds extend along geodesics. Specifically, the [Local Isometry from Complete Manifold is a Covering Map](/theorems/2735) theorem says: a local isometry from a complete connected Riemannian manifold to a connected Riemannian manifold is a Riemannian covering map. Compact manifolds are complete (by [Hopf–Rinow Theorem](/theorems/2726)), so $F$ is a covering. **Why a covering of a torus is itself a torus.** Universal covers of flat tori are $\mathbb{R}^n$, which has zero fundamental group. Pulling back the covering $F: M \to T^n$ to universal covers gives $\widetilde{M} \cong \mathbb{R}^n$ (Riemannian isometry), and $M$ is the quotient by a deck-transformation group acting freely and properly discontinuously by isometries. The classification of such groups says they are full-rank lattices of translations — Bieberbach's theorem in full generality is not needed because the deck transformations descend from $\Lambda$ via the covering of tori, hence are translations. **Why the rank of $\Lambda$ is exactly $n$, not less.** A naive concern: what if the period map produces a lattice of lower rank, so that $\mathbb{R}^n / \Lambda$ is a cylinder rather than a compact torus? The answer is that the de Rham theorem pairs $H^1_{\mathrm{dR}}(M; \mathbb{R})$ with $H_1(M; \mathbb{R})$ non-degenerately, and $\dim H^1_{\mathrm{dR}}(M; \mathbb{R}) = b^1(M) = n$ by hypothesis. So the period map $H_1(M; \mathbb{R}) \to \mathbb{R}^n$ is a linear isomorphism, and the period image $\Lambda \otimes \mathbb{R}$ is all of $\mathbb{R}^n$. Discreteness combined with finite generation forces $\Lambda$ to be a lattice of rank $n$. **The dim-equal-$b^1$ condition is the heart of the rigidity.** Without $b^1(M) = n$, parts (1) and (2) of the Bochner theorem still give us at most $n$ parallel $1$-forms but possibly fewer. A non-saturated case ($b^1(M) < n$) does not produce a global parallel coframe, hence no map to a torus. The saturation $b^1(M) = n$ is what unlocks all the geometry. [/guided] [/step]