[proofplan] We prove the four parts in order. Part 1 applies the [Maximal Euclidean Factor](/theorems/2768) to the universal cover $\tilde M$, giving a splitting $\tilde M \cong X \times \mathbb R^q$ with $X$ line-free; the remaining content (compactness and simple connectedness of $X$) is outlined and cited from the standard references — Cheeger–Ebin and Petersen — because it depends on Bieberbach-style results about cocompact discrete actions on $\mathbb R^q$. Part 2 uses the contrapositive of [Universal Cover of Compact Manifold with Nonneg Ricci](/theorems/2770) together with the line-splitting structure: if $\pi_1(M)$ were infinite, $\tilde M$ would split off an $\mathbb R$ factor whose generator has zero Ricci in its own direction, contradicting strict positivity at $p$. Part 3 is the standard de Rham-style fact that the isometry group of a Riemannian product splits as the product of factor isometry groups when the factors are appropriately rigid; the $\mathbb R^q$ factor's isometry group is the Euclidean motion group $E(\mathbb R^q) = O(q) \ltimes \mathbb R^q$. Part 4 follows from part 3 by projecting a transitive action on $\tilde M = X \times \mathbb R^q$ to the first factor. This proof presents parts 1 and 3 as outlines with references to standard texts; the technical inputs (Bieberbach's theorem, the de Rham splitting of isometry groups) are treated as black boxes. Parts 2 and 4 are given in full rigor. [/proofplan] [step:Prove part 1: split $\tilde M$ as $X \times \mathbb R^q$ with $X$ compact and simply connected] Apply [Maximal Euclidean Factor](/theorems/2768) to $(\tilde M, \tilde g)$. The hypotheses of that theorem are: - $\tilde M$ is complete: this follows from compactness of $M$ via Hopf–Rinow lifted along the local isometry $\pi: \tilde M \to M$ (see [Non-compact Universal Cover Contains a Line](/theorems/2769) for the explicit lifting argument). - $\tilde M$ is connected: $\tilde M$ is path-connected (as a covering space of the connected $M$, it is the universal cover, in particular connected). - $\operatorname{Ric}(\tilde g) \geq 0$: $\pi$ is a local isometry, hence preserves the Ricci tensor pointwise: $\operatorname{Ric}(\tilde g)_{\tilde p} = \pi^* \operatorname{Ric}(g)_{\pi(\tilde p)}$. Since $\operatorname{Ric}(g) \geq 0$, the same holds for $\tilde g$. Theorem 2768 produces an isometry \begin{align*} (\tilde M, \tilde g) \cong (X \times \mathbb R^q,\; g_X + g_{\mathrm{Eucl}}) \end{align*} where $(X, g_X)$ is complete, connected, has $\operatorname{Ric}(g_X) \geq 0$, and contains no line, with $X$ and $q$ uniquely determined. It remains to show $X$ is compact and simply connected. **Simple connectedness of $X$.** $\tilde M = X \times \mathbb R^q$ is simply connected (as the universal cover of $M$). Apply the Künneth-type formula for $\pi_1$ of a product of pointed spaces: \begin{align*} \pi_1(X \times \mathbb R^q) = \pi_1(X) \times \pi_1(\mathbb R^q). \end{align*} Since $\mathbb R^q$ is simply connected ($\pi_1(\mathbb R^q) = 0$), we get $\pi_1(X) \cong \pi_1(\tilde M) = 0$. Hence $X$ is simply connected. **Compactness of $X$ — outline.** This is the technically difficult part. We outline the standard argument; see Cheeger–Ebin, *Comparison and Applications of Analytic and Geometric Methods in Riemannian Geometry* (or the modern reformulation in Petersen, *Riemannian Geometry*, Chapter 9), for the full proof. The deck-transformation group $\Gamma = \pi_1(M)$ acts on $\tilde M = X \times \mathbb R^q$ by isometries with compact quotient $M$. By part 3 (proved below independently of part 1's compactness assertion), every isometry of $X \times \mathbb R^q$ is of the form $(\phi, \psi)$ with $\phi \in I(X)$ and $\psi \in E(\mathbb R^q)$. The projection \begin{align*} p_2: \Gamma \to E(\mathbb R^q), \quad \gamma = (\phi, \psi) \mapsto \psi \end{align*} has image $\Gamma_2 \subseteq E(\mathbb R^q)$, a discrete group of Euclidean motions. By Bieberbach's first theorem on crystallographic groups, the image $\Gamma_2$ contains a finite-index subgroup of pure translations isomorphic to $\mathbb Z^{q'}$ for some $q' \leq q$; cocompactness of $\Gamma$ on $\tilde M$ forces $q' = q$, so $\Gamma_2$ is a cocompact discrete subgroup of $E(\mathbb R^q)$ — a so-called Bieberbach group. The kernel $\Gamma_1 := \ker(p_2) \subseteq I(X)$ then acts on $X$ — and by the cocompactness statement on $\tilde M$ together with the cocompactness of $\Gamma_2$ on $\mathbb R^q$, the action of $\Gamma_1$ on $X$ has compact quotient. So $\Gamma_1 \backslash X$ is compact. If $X$ were non-compact, then by [Non-compact Universal Cover Contains a Line](/theorems/2769) applied to the cover $X \to \Gamma_1 \backslash X$ (compact), $X$ would contain a line — but $X$ contains no line by construction (it is the maximal-Euclidean-factor base from theorem 2768). Hence $X$ is compact. [guided] We apply [Maximal Euclidean Factor](/theorems/2768) to the universal Riemannian cover $(\tilde M, \tilde g)$. Why is this the right tool? Theorem 2768 produces the canonical splitting of any complete connected Riemannian manifold of non-negative Ricci curvature into a "no-line" factor times a Euclidean factor — exactly the conclusion we want for part 1. We must verify its three hypotheses. *Completeness of $\tilde M$.* Since $M$ is compact, it is geodesically complete by Hopf–Rinow. The covering map $\pi: \tilde M \to M$ is a local isometry, hence lifts geodesics — every maximal geodesic in $\tilde M$ projects to a geodesic in $M$, and the projected geodesic exists for all time, so the lifted one does as well. Thus $\tilde M$ is complete (this is the lifting argument made explicit in [Non-compact Universal Cover Contains a Line](/theorems/2769)). *Connectedness of $\tilde M$.* The universal cover of a path-connected space is path-connected by construction. *Non-negative Ricci on $\tilde g$.* Since $\pi$ is a local isometry, the Ricci tensor pulls back pointwise: $\operatorname{Ric}(\tilde g)_{\tilde p} = \pi^* \operatorname{Ric}(g)_{\pi(\tilde p)}$ at every $\tilde p \in \tilde M$. With $\operatorname{Ric}(g) \geq 0$ on $M$, this gives $\operatorname{Ric}(\tilde g) \geq 0$ on $\tilde M$. All three hypotheses hold. Theorem 2768 then produces an isometry \begin{align*} (\tilde M, \tilde g) \cong (X \times \mathbb R^q,\; g_X + g_{\mathrm{Eucl}}) \end{align*} with $(X, g_X)$ complete, connected, non-negative Ricci, and *line-free*; the integers $q \geq 0$ and the Riemannian manifold $X$ (up to isometry) are uniquely determined by $\tilde M$. The line-free property of $X$ is the defining feature: $X$ is what remains after stripping off every Euclidean direction. To finish part 1 we must show $X$ is compact and simply connected. *Simple connectedness.* The universal cover $\tilde M$ is simply connected by definition. We use the product formula for the fundamental group: for pointed spaces $(A, a_0)$ and $(B, b_0)$, \begin{align*} \pi_1(A \times B,\, (a_0, b_0)) \cong \pi_1(A, a_0) \times \pi_1(B, b_0). \end{align*} Applied to $\tilde M = X \times \mathbb R^q$, and using $\pi_1(\mathbb R^q) = 0$ (since $\mathbb R^q$ is contractible), \begin{align*} 0 = \pi_1(\tilde M) = \pi_1(X) \times \pi_1(\mathbb R^q) = \pi_1(X) \times 0 \cong \pi_1(X). \end{align*} So $X$ is simply connected. *Compactness — outline.* This is the technically delicate part, and we sketch the standard argument; see Cheeger–Ebin, *Comparison and Applications of Analytic and Geometric Methods in Riemannian Geometry*, or Petersen, *Riemannian Geometry*, 3rd edition, Chapter 9, for the full proof. The strategy is to use the deck-transformation action of $\Gamma := \pi_1(M)$ on $\tilde M = X \times \mathbb R^q$. Since $M = \Gamma \backslash \tilde M$ is compact, $\Gamma$ acts cocompactly on $\tilde M$ by isometries. By part 3 (proved below independently of compactness), every isometry of $X \times \mathbb R^q$ splits as $(\phi, \psi)$ with $\phi \in I(X)$ and $\psi \in E(\mathbb R^q)$. So we get a homomorphism \begin{align*} p_2: \Gamma \to E(\mathbb R^q), \qquad \gamma = (\phi, \psi) \mapsto \psi \end{align*} whose image $\Gamma_2 := p_2(\Gamma)$ is a discrete subgroup of $E(\mathbb R^q)$. Why discrete? Because $\Gamma$ acts properly discontinuously on $\tilde M$, and the projection of a properly discontinuous action onto the second factor is properly discontinuous, hence discrete. We claim $\Gamma_2$ is cocompact in $E(\mathbb R^q)$. Cocompactness of $\Gamma$ on $\tilde M$ together with the line-free condition on $X$ forces this: orbits of $\Gamma$ in $\tilde M$ project to cocompact orbits of $\Gamma_2$ in $\mathbb R^q$ (any non-cocompactness would manifest as a missing direction in $\mathbb R^q$, contradicting maximality of the Euclidean factor in theorem 2768). By Bieberbach's first theorem on crystallographic groups, every cocompact discrete subgroup of $E(\mathbb R^q) = O(q) \ltimes \mathbb R^q$ contains a finite-index subgroup of pure translations isomorphic to $\mathbb Z^q$ (full rank in $\mathbb R^q$). So $\Gamma_2$ is virtually $\mathbb Z^q$, a Bieberbach group. The kernel $\Gamma_1 := \ker(p_2) \subseteq I(X)$ acts on $X$. Cocompactness of $\Gamma$ on $\tilde M = X \times \mathbb R^q$, combined with cocompactness of $\Gamma_2$ on $\mathbb R^q$, gives that $\Gamma_1$ acts cocompactly on $X$, so $\Gamma_1 \backslash X$ is compact. Now the line-free property of $X$ enters. If $X$ were non-compact, then by [Non-compact Universal Cover Contains a Line](/theorems/2769) applied to the cover $X \to \Gamma_1 \backslash X$ — non-compact total space, compact quotient, both with $\operatorname{Ric} \geq 0$ — $X$ would contain a line. But $X$ is line-free by construction of the maximal-Euclidean-factor decomposition. Contradiction. Hence $X$ is compact. We do not present full details here because (a) Bieberbach's theorem is a standalone classical result whose self-contained proof would dominate the present argument, and (b) the structure of compact-quotient actions on Riemannian products is a standard consequence of the de Rham decomposition. The classical reference is Cheeger–Gromoll's original 1971 paper *The structure of complete manifolds of nonnegative curvature*; the modern textbook treatment is Petersen, *Riemannian Geometry*, 3rd edition, Chapter 9 (especially §9.2.5–9.3 on the structure of isometries) and Chapter 12 (de Rham's theorem). [/guided] [/step] [step:Prove part 2: $\operatorname{Ric}(g)_p > 0$ at one point implies $\pi_1(M)$ finite] We argue by contradiction. Suppose $\pi_1(M)$ is infinite. We will derive a contradiction with the hypothesis $\operatorname{Ric}(g)_p > 0$. Apply [Universal Cover of Compact Manifold with Nonneg Ricci](/theorems/2770). Its hypotheses are: - $(M, g)$ compact connected Riemannian: by hypothesis of the present theorem. - $\operatorname{Ric}(g) \geq 0$: by hypothesis. - $\pi_1(M)$ infinite: by our assumption-for-contradiction. Theorem 2770 yields a line $\ell: \mathbb R \to \tilde M$ in the universal cover $\tilde M$. Apply [Cheeger–Gromoll Line-Splitting Theorem](/theorems/2767) to $(\tilde M, \tilde g)$ with the line $\ell$. Hypotheses: - $\tilde M$ complete: as in part 1. - $\operatorname{Ric}(\tilde g) \geq 0$: as in part 1, by local isometry. - $\tilde M$ contains a line: just established. Theorem 2767 produces an isometry \begin{align*} (\tilde M, \tilde g) \cong (N \times \mathbb R, g_N + dt^2) \end{align*} for some complete Riemannian manifold $(N, g_N)$ with $\operatorname{Ric}(g_N) \geq 0$. The $\partial_t$ vector field on the $\mathbb R$ factor pulls back to a unit-norm parallel vector field $V \in \Gamma(T \tilde M)$ on $\tilde M$, with $V_{\tilde q}$ tangent to the $\mathbb R$-direction at every $\tilde q \in \tilde M$. [claim:If $V \in \Gamma(T \tilde M)$ is a parallel vector field on $\tilde M$, then $\operatorname{Ric}(\tilde g)(V, V) \equiv 0$ on $\tilde M$] [proof] A parallel vector field satisfies $\nabla V = 0$, where $\nabla$ is the Levi-Civita connection on $\tilde M$. We compute the Ricci curvature in the direction of $V$. The Ricci tensor on $\tilde M = N \times \mathbb R$ decomposes as a direct sum on the product tangent space: at a point $(q, t) \in N \times \mathbb R$, \begin{align*} T_{(q, t)}\tilde M = T_q N \oplus T_t \mathbb R, \qquad \operatorname{Ric}(\tilde g)_{(q, t)} = \operatorname{Ric}(g_N)_q \oplus \operatorname{Ric}(dt^2)_t. \end{align*} The Euclidean line $(\mathbb R, dt^2)$ is flat, so $\operatorname{Ric}(dt^2) = 0$ on $\mathbb R$. The vector $V_{(q, t)} = \partial_t \in T_t \mathbb R$ lies in the $\mathbb R$-summand, where the Ricci form is zero. Hence \begin{align*} \operatorname{Ric}(\tilde g)_{(q, t)}(V_{(q, t)}, V_{(q, t)}) = \operatorname{Ric}(dt^2)_t(\partial_t, \partial_t) = 0 \end{align*} identically. [/proof] [/claim] Now project to $M$. Pick any $\tilde p \in \pi^{-1}(p)$ (a lift of the strict-positivity point). Local-isometry preservation of the Ricci tensor: \begin{align*} \operatorname{Ric}(\tilde g)_{\tilde p}(V_{\tilde p}, V_{\tilde p}) = \operatorname{Ric}(g)_p(d\pi_{\tilde p} V_{\tilde p}, d\pi_{\tilde p} V_{\tilde p}). \end{align*} The LHS is $0$ by the claim. The RHS is the Ricci tensor at $p$ evaluated on the non-zero vector $d\pi_{\tilde p} V_{\tilde p} \in T_p M$ (non-zero because $|V_{\tilde p}|_{\tilde g} = 1$ and $d\pi_{\tilde p}$ is a linear isometry, so $|d\pi_{\tilde p} V_{\tilde p}|_g = 1$). By the strict-positivity hypothesis $\operatorname{Ric}(g)_p > 0$ — meaning $\operatorname{Ric}(g)_p$ is positive definite as a symmetric bilinear form on $T_p M$ — we have \begin{align*} \operatorname{Ric}(g)_p(w, w) > 0 \quad \text{for every nonzero } w \in T_p M. \end{align*} In particular, with $w = d\pi_{\tilde p} V_{\tilde p} \neq 0$, the RHS is strictly positive, contradicting the LHS being zero. The contradiction shows our assumption $|\pi_1(M)| = \infty$ was false. Hence $\pi_1(M)$ is finite. [guided] The strategy is contradiction. Suppose $\pi_1(M)$ is *infinite*. We will produce a non-zero tangent vector at $p$ on which $\operatorname{Ric}(g)_p$ vanishes — contradicting strict positivity. *Why does standard Myers not work?* Myers' theorem gives finiteness of $\pi_1$ from a *uniform* lower bound $\operatorname{Ric} \geq (n-1)k > 0$ globally, by deriving a diameter bound and then compactness of $\tilde M$. Here we have only $\operatorname{Ric}(g)_p > 0$ at a *single* point — no uniform bound — so Myers' diameter estimate fails. The line-splitting argument is strictly more powerful: it derives a contradiction from strict positivity at one single point. *Step 1: produce a line in $\tilde M$.* We invoke [Universal Cover of Compact Manifold with Nonneg Ricci](/theorems/2770), which states: if $(M, g)$ is compact connected with $\operatorname{Ric}(g) \geq 0$ and $\pi_1(M)$ is *infinite*, then the universal cover $\tilde M$ contains a line. We verify the hypotheses: - $(M, g)$ compact connected Riemannian — given by the theorem statement. - $\operatorname{Ric}(g) \geq 0$ — given by the theorem statement. - $\pi_1(M)$ infinite — our assumption-for-contradiction. All three hold, so theorem 2770 produces a line $\ell: \mathbb R \to \tilde M$ in $\tilde M$ — i.e., a unit-speed geodesic that is globally distance-minimizing on every interval $[a, b]$. *Step 2: split off an $\mathbb R$ factor.* We invoke the [Cheeger–Gromoll Line-Splitting Theorem](/theorems/2767) on $(\tilde M, \tilde g)$ with the line $\ell$. Hypotheses: - $\tilde M$ complete: $M$ is compact, hence complete by Hopf–Rinow; $\pi: \tilde M \to M$ is a local isometry, so geodesics in $\tilde M$ project to geodesics in $M$ that exist for all time, and so do their lifts. Hence $\tilde M$ is complete. - $\operatorname{Ric}(\tilde g) \geq 0$: $\pi$ is a local isometry, so $\operatorname{Ric}(\tilde g)_{\tilde q} = \pi^* \operatorname{Ric}(g)_{\pi(\tilde q)} \geq 0$ pointwise. - $\tilde M$ contains a line: just established in step 1. Theorem 2767 then yields an isometric splitting \begin{align*} (\tilde M, \tilde g) \cong (N \times \mathbb R, g_N + dt^2) \end{align*} for some complete Riemannian manifold $(N, g_N)$ with $\operatorname{Ric}(g_N) \geq 0$. The $\partial_t$-vector field on the $\mathbb R$ factor pulls back to a unit-norm parallel vector field $V \in \Gamma(T \tilde M)$, defined globally on $\tilde M$, with $V_{\tilde q}$ tangent to the $\mathbb R$-direction at every $\tilde q$. *Step 3: parallel vector fields on a Riemannian product have zero Ricci in their own direction.* We compute $\operatorname{Ric}(\tilde g)(V, V)$. The Ricci tensor on a Riemannian product is block-diagonal: at any point $(q, t) \in N \times \mathbb R$, the tangent space splits as $T_{(q, t)}\tilde M = T_q N \oplus T_t \mathbb R$, and \begin{align*} T_{(q, t)}\tilde M = T_q N \oplus T_t \mathbb R, \qquad \operatorname{Ric}(\tilde g)_{(q, t)} = \operatorname{Ric}(g_N)_q \oplus \operatorname{Ric}(dt^2)_t. \end{align*} The second summand is the Ricci tensor of $(\mathbb R, dt^2)$, which is *flat* (one-dimensional Euclidean space has zero curvature), hence $\operatorname{Ric}(dt^2) \equiv 0$. Since $V_{(q, t)} = \partial_t \in T_t \mathbb R$ lives entirely in the $\mathbb R$-summand, \begin{align*} \operatorname{Ric}(\tilde g)_{(q, t)}(V_{(q, t)}, V_{(q, t)}) = \operatorname{Ric}(dt^2)_t(\partial_t, \partial_t) = 0 \end{align*} identically on $\tilde M$. (This is the content of the boxed claim above — it is the precise reason a flat factor "kills" Ricci in its own direction.) *Step 4: project to $p$ and find the contradiction.* Pick any lift $\tilde p \in \pi^{-1}(p)$. Since $\pi$ is a local isometry, $d\pi_{\tilde p}: T_{\tilde p}\tilde M \to T_p M$ is a linear isometry, so it preserves bilinear forms: \begin{align*} \operatorname{Ric}(\tilde g)_{\tilde p}(V_{\tilde p}, V_{\tilde p}) = \operatorname{Ric}(g)_p(d\pi_{\tilde p} V_{\tilde p}, d\pi_{\tilde p} V_{\tilde p}). \end{align*} The LHS is $0$ by step 3. The RHS evaluates the Ricci tensor at $p$ on the vector $w := d\pi_{\tilde p} V_{\tilde p} \in T_p M$. Note $w \neq 0$: $V_{\tilde p}$ has $\tilde g$-norm 1, and $d\pi_{\tilde p}$ is a *linear isometry* (not merely linear), so $|w|_g = |V_{\tilde p}|_{\tilde g} = 1$. Now we consume the hypothesis $\operatorname{Ric}(g)_p > 0$ — meaning $\operatorname{Ric}(g)_p$ is positive definite as a symmetric bilinear form on the *whole* tangent space $T_p M$. Positive definiteness gives \begin{align*} \operatorname{Ric}(g)_p(w, w) > 0 \quad \text{for every nonzero } w \in T_p M. \end{align*} With our specific $w = d\pi_{\tilde p} V_{\tilde p} \neq 0$, the RHS is strictly positive — contradicting LHS $= 0$. The strict positivity at a *single* point is consumed in exactly this place: we needed $\operatorname{Ric}(g)_p$ to be positive on *every* nonzero vector of $T_p M$, because we did not get to choose the direction $w$ — it was forced on us by the line-splitting. This is why "strict positivity at one point" is the exact right hypothesis: positive definiteness is what makes the argument robust to the unknown direction of the line. *Conclusion.* Our assumption $|\pi_1(M)| = \infty$ has produced a contradiction. Hence $\pi_1(M)$ is finite. [/guided] [/step] [step:Prove part 3: $I(\tilde M) = I(X) \times E(\mathbb R^q)$] We outline the standard proof, citing the de Rham decomposition theorem; full details are in Kobayashi–Nomizu, *Foundations of Differential Geometry, Volume 1*, Chapter 4. Recall from part 1 that $(\tilde M, \tilde g) \cong (X \times \mathbb R^q, g_X + g_{\mathrm{Eucl}})$ as Riemannian manifolds, with $X$ simply connected, $\operatorname{Ric}(g_X) \geq 0$, and $X$ containing no line. The de Rham decomposition theorem asserts: for a simply connected Riemannian manifold $\tilde M = M_1 \times \cdots \times M_k$ given as a Riemannian product where each $M_i$ has its tangent space at every point spanned by an irreducible holonomy representation, every isometry of $\tilde M$ permutes the factors and decomposes as a product of isometries of factors. In our setting the decomposition has only two distinguished factors — $X$ and $\mathbb R^q$ — separated by the property of "containing a line" (the $\mathbb R^q$ factor is determined up to isometry as the maximal flat factor of $\tilde M$). [claim:$X$ is not isometric to a Riemannian product with any non-trivial Euclidean factor] [proof] By construction (theorem 2768), $X$ is the maximal-line-free factor of $\tilde M$: $X$ contains no line. If $X = X' \times \mathbb R$ were a Riemannian product with a non-trivial flat factor, then the curve $\{x_0'\} \times \mathbb R \subset X' \times \mathbb R = X$ would be a unit-speed geodesic with $d_X((x_0', s), (x_0', t)) = |s - t|$, i.e., a line in $X$. This contradicts $X$ being line-free. Hence $X$ has no flat factor splitting off. [/proof] [/claim] By the claim, the decomposition $\tilde M = X \times \mathbb R^q$ separates a "line-free part" $X$ from a maximal flat part $\mathbb R^q$. Any isometry $f: \tilde M \to \tilde M$ must respect this separation: it cannot send a non-flat direction (in $X$) to a flat direction (in $\mathbb R^q$) or vice versa, because both are intrinsically defined via parallel transport and curvature properties. More precisely, the de Rham decomposition gives: \begin{align*} I(X \times \mathbb R^q) = I(X) \times I(\mathbb R^q), \end{align*} with each isometry $f$ written uniquely as $f(x, y) = (f_1(x), f_2(y))$ for $f_1 \in I(X)$, $f_2 \in I(\mathbb R^q)$. The isometry group of Euclidean space $(\mathbb R^q, g_{\mathrm{Eucl}})$ is the rigid motion group $E(\mathbb R^q) = O(q) \ltimes \mathbb R^q$: every isometry is of the form $\mathbf y \mapsto A\mathbf y + \mathbf b$ for some $A \in O(q)$ and $\mathbf b \in \mathbb R^q$. This is a standard linear-algebra fact about $\mathbb R^q$: an isometry of Euclidean space is the composition of a translation and an orthogonal transformation. Combining, \begin{align*} I(\tilde M) = I(X \times \mathbb R^q) = I(X) \times E(\mathbb R^q), \end{align*} which is the assertion of part 3. [guided] The strategy is to apply the de Rham decomposition theorem to the splitting $\tilde M = X \times \mathbb R^q$ from part 1. The de Rham theorem says: a simply connected complete Riemannian manifold that splits as a Riemannian product with "incompatible" factors has its isometry group splitting as the product of factor isometry groups. We must verify the incompatibility, then identify the isometry group of each factor. *De Rham decomposition theorem (statement in our setting).* Let $\tilde M$ be simply connected and complete, and suppose $\tilde M = M_1 \times M_2$ as a Riemannian product. If the two factors $M_1, M_2$ are determined up to isometry by intrinsic invariants of $\tilde M$ (concretely, by the holonomy representation, which must split as a direct sum of representations that do not share an irreducible summand), then every isometry $f \in I(\tilde M)$ has the form $f = (f_1, f_2)$ with $f_1 \in I(M_1)$ and $f_2 \in I(M_2)$, and consequently \begin{align*} I(M_1 \times M_2) = I(M_1) \times I(M_2). \end{align*} The full proof is in Kobayashi–Nomizu, *Foundations of Differential Geometry, Vol. 1*, Chapter 4; we treat the de Rham theorem as a black box. *Verifying the hypotheses for $\tilde M = X \times \mathbb R^q$.* From part 1: $\tilde M$ is simply connected (universal cover of $M$) and complete (Hopf–Rinow lifted). The product structure is given. We must verify the "incompatibility" of the two factors $X$ and $\mathbb R^q$. [claim:$X$ is not isometric to a Riemannian product with any non-trivial Euclidean factor] [proof] By construction (theorem 2768), $X$ is the maximal-line-free factor of $\tilde M$: $X$ contains no line. Suppose for contradiction $X \cong X' \times \mathbb R$ as Riemannian product, with non-trivial flat factor $\mathbb R$. Pick any $x_0' \in X'$. The slice \begin{align*} \gamma: \mathbb R \to X, \qquad t \mapsto (x_0', t) \end{align*} is a unit-speed geodesic in $X$ (geodesics in a Riemannian product are products of geodesics, and $t \mapsto t$ is a unit-speed geodesic in $(\mathbb R, dt^2)$). Moreover, on a Riemannian product, the distance function decomposes as $d_{X' \times \mathbb R}((a, s), (b, t))^2 = d_{X'}(a, b)^2 + |s - t|^2$, so \begin{align*} d_X((x_0', s), (x_0', t)) = \sqrt{0 + |s - t|^2} = |s - t|. \end{align*} Hence $\gamma$ is a line in $X$ — contradicting the line-free property of $X$. So $X$ has no flat factor splitting off. [/proof] [/claim] The claim says the flat directions of $\tilde M$ are *exactly* concentrated in the $\mathbb R^q$ factor — none hide inside $X$. This is the source of the holonomy incompatibility: the holonomy representation of $\mathbb R^q$ at any point is trivial (Euclidean space has trivial holonomy), and it captures all the parallel directions of $\tilde M$. Any parallel direction in $X$ would, by the same argument as the claim, generate a line in $X$ — but $X$ has none. So the holonomy representations of $X$ and $\mathbb R^q$ share no irreducible summand: they are incompatible in the precise sense the de Rham theorem needs. *Applying de Rham.* Hypotheses verified, the de Rham theorem yields \begin{align*} I(X \times \mathbb R^q) = I(X) \times I(\mathbb R^q), \end{align*} with each isometry $f$ of $\tilde M$ written uniquely as $f(x, y) = (f_1(x), f_2(y))$ for $f_1 \in I(X)$, $f_2 \in I(\mathbb R^q)$. *Identifying $I(\mathbb R^q)$.* It remains to show $I(\mathbb R^q) = E(\mathbb R^q) = O(q) \ltimes \mathbb R^q$. Let $f: \mathbb R^q \to \mathbb R^q$ be an isometry of Euclidean space. Define $A := df_0$, the differential of $f$ at the origin. Since $f$ preserves the Euclidean inner product, $A \in O(q)$ is a linear isometry. We claim $f(\mathbf y) = A \mathbf y + f(0)$ for all $\mathbf y$. The map $g(\mathbf y) := f(\mathbf y) - f(0)$ is also an isometry of $\mathbb R^q$ (Euclidean translations are isometries) and fixes the origin: $g(0) = 0$. An isometry of Euclidean space fixing the origin is linear (this is the Mazur–Ulam theorem in its classical form), so $g(\mathbf y) = dg_0 \mathbf y = A \mathbf y$. Setting $\mathbf b := f(0)$, we obtain \begin{align*} f(\mathbf y) = A \mathbf y + \mathbf b, \qquad A \in O(q), \; \mathbf b \in \mathbb R^q. \end{align*} This identification gives $I(\mathbb R^q) \cong E(\mathbb R^q) = O(q) \ltimes \mathbb R^q$ as the *Euclidean motion group*. The semidirect-product structure comes from composing two such maps: \begin{align*} (A_1, \mathbf b_1) \cdot (A_2, \mathbf b_2): \mathbf z \mapsto A_1(A_2 \mathbf z + \mathbf b_2) + \mathbf b_1 = A_1 A_2 \mathbf z + (A_1 \mathbf b_2 + \mathbf b_1), \end{align*} i.e., $(A_1, \mathbf b_1)(A_2, \mathbf b_2) = (A_1 A_2, A_1 \mathbf b_2 + \mathbf b_1)$, the standard $O(q) \ltimes \mathbb R^q$ multiplication. *Combining.* Substituting into the de Rham splitting, \begin{align*} I(\tilde M) = I(X \times \mathbb R^q) = I(X) \times E(\mathbb R^q), \end{align*} which is the assertion of part 3. [/guided] [/step] [step:Prove part 4: $\tilde M$ homogeneous implies $X$ homogeneous] Suppose $\tilde M = X \times \mathbb R^q$ is homogeneous, that is, $I(\tilde M)$ acts transitively on $\tilde M$. By part 3, every isometry $f \in I(\tilde M)$ has the form $f(x, y) = (f_1(x), f_2(y))$ with $f_1 \in I(X)$ and $f_2 \in E(\mathbb R^q)$. Fix any two points $x_1, x_2 \in X$. Pick any $y_0 \in \mathbb R^q$. By transitivity of $I(\tilde M)$ on $\tilde M$, there exists $f \in I(\tilde M)$ with \begin{align*} f(x_1, y_0) = (x_2, y_0). \end{align*} Writing $f(x, y) = (f_1(x), f_2(y))$, this says $f_1(x_1) = x_2$ and $f_2(y_0) = y_0$ (the latter is irrelevant for our purpose). Hence $f_1 \in I(X)$ sends $x_1$ to $x_2$. Since $x_1, x_2 \in X$ were arbitrary, $I(X)$ acts transitively on $X$, so $X$ is homogeneous. [guided] The strategy is a direct projection. We are given that $\tilde M = X \times \mathbb R^q$ is homogeneous — its isometry group $I(\tilde M)$ acts transitively on $\tilde M$. We want to deduce that $I(X)$ acts transitively on $X$. The bridge is part 3: every isometry of $\tilde M$ splits as a pair $(f_1, f_2)$ of factor isometries, and the first component will give us transitivity on $X$. *Setup.* Fix any two points $x_1, x_2 \in X$. We must produce an isometry $f_1 \in I(X)$ with $f_1(x_1) = x_2$. Pick an arbitrary auxiliary point $y_0 \in \mathbb R^q$ — its specific value does not matter. Form the two points $(x_1, y_0), (x_2, y_0) \in X \times \mathbb R^q = \tilde M$. *Apply transitivity on $\tilde M$.* By the hypothesis that $\tilde M$ is homogeneous, $I(\tilde M)$ acts transitively, so there exists an isometry $f \in I(\tilde M)$ with \begin{align*} f(x_1, y_0) = (x_2, y_0). \end{align*} *Apply part 3 to decompose $f$.* By part 3 (proved above), every isometry of $\tilde M = X \times \mathbb R^q$ has the form \begin{align*} f(x, y) = (f_1(x), f_2(y)) \quad \text{for some } f_1 \in I(X), \; f_2 \in E(\mathbb R^q). \end{align*} This is the key input: without part 3, $f$ could in principle mix the factors (e.g., send $X$-directions into $\mathbb R^q$-directions), and projection onto the first factor would not yield an isometry. Part 3 guarantees the splitting. *Project onto the $X$-factor.* Equating the two expressions for $f(x_1, y_0)$: \begin{align*} (f_1(x_1), f_2(y_0)) = f(x_1, y_0) = (x_2, y_0). \end{align*} Comparing first components, $f_1(x_1) = x_2$. (We do not need to use the second-component equality $f_2(y_0) = y_0$; in fact, even if we had only required $f(x_1, y_0) = (x_2, y_0')$ for some $y_0' \in \mathbb R^q$, the first-component projection would still give $f_1(x_1) = x_2$.) *Conclusion.* Since $x_1, x_2 \in X$ were arbitrary, for every pair of points in $X$ there exists an isometry $f_1 \in I(X)$ sending one to the other. Hence $I(X)$ acts transitively on $X$, i.e., $X$ is homogeneous. This is an instance of a general phenomenon: when a transitive group action on a product space splits as a product of factor actions, each factor action is itself transitive. The non-trivial input is that the action *does* split — and that is exactly the content of part 3. [/guided] [/step]