[proofplan] The argument has three layers. First, we observe that $\mathrm{Hol}^0(M)$, the **restricted holonomy group** at any point $x$, is by definition the image under parallel transport of null-homotopic loops at $x$. Second, we adapt the proof of [Simple Connectivity and Path Connectivity](/theorems/2762) to the restricted setting: the based subspace of null-homotopic loops $\Omega^0(x, x) \subseteq \Omega(x, x)$ is always path-connected (independent of $\pi_1(M)$), because every null-homotopic loop has a continuous null-homotopy to the constant loop, and continuous deformations through null-homotopic loops connect any two such loops to the constant loop. Continuity of the parallel transport map $P : \Omega(x, x) \to \mathrm{O}(n)$ — established in the same theorem — then makes $\mathrm{Hol}^0(M) = P(\Omega^0(x, x))$ a path-connected subgroup of $\mathrm{O}(n)$. Third, we use the topological fact that $\mathrm{O}(n)$ has exactly two path components, namely $\mathrm{SO}(n)$ (determinant $+1$) and the coset $\mathrm{O}(n) \setminus \mathrm{SO}(n)$ (determinant $-1$), distinguished by the continuous homomorphism $\det : \mathrm{O}(n) \to \{\pm 1\}$. A path-connected subgroup containing the identity $\mathrm{Id} \in \mathrm{SO}(n)$ must lie entirely in the path component of the identity, namely $\mathrm{SO}(n)$. [/proofplan]

[step:Set up the restricted holonomy group as the image of null-homotopic loops] Let $(M, g)$ be a connected Riemannian manifold of dimension $n$, $\nabla$ its Levi-Civita connection, and fix $x \in M$. By the [Definition of the Restricted Holonomy Group](/page/Restricted%20Holonomy%20Group), \begin{align*} \mathrm{Hol}^0_x(M) := \{P_\gamma : \gamma \in \Omega^0(x, x)\} \subseteq \mathrm{O}(T_x M), \end{align*} where $\Omega^0(x, x) \subseteq \Omega(x, x)$ is the subspace of piecewise-smooth loops at $x$ that are **null-homotopic** in $M$ — that is, there exists a continuous map $H : [0, 1]^2 \to M$ with $H(0, \cdot) = c_x$ (the constant loop), $H(1, \cdot) = \gamma$, and $H(s, 0) = H(s, 1) = x$ for every $s$. After choosing a $g_x$-orthonormal basis of $T_x M$, we identify $\mathrm{O}(T_x M) \cong \mathrm{O}(n)$, and write \begin{align*} P : \Omega(x, x) \to \mathrm{O}(n), \qquad \mathrm{Hol}^0_x(M) = P(\Omega^0(x, x)). \end{align*} The full holonomy group $\mathrm{Hol}_x(M) = P(\Omega(x, x))$ may depend on $x$ (up to conjugation by parallel transport along a path connecting two points), but the restricted holonomy at different points are conjugate, so we drop the basepoint and write $\mathrm{Hol}^0(M)$ when convenient — the statement $\mathrm{Hol}^0(M) \subseteq \mathrm{SO}(n)$ is invariant under conjugation by elements of $\mathrm{O}(n)$, so the choice of basepoint and orthonormal basis is immaterial. The topology on $\Omega(x, x)$ is the topology of uniform convergence of paths and velocities on the smooth pieces, as in the proof of [Simple Connectivity and Path Connectivity](/theorems/2762). The subspace $\Omega^0(x, x)$ inherits the subspace topology. [/step]

[step:Show that $\Omega^0(x, x)$ is path-connected for any connected $M$] We claim: for any connected Riemannian manifold $M$, the space $\Omega^0(x, x)$ of null-homotopic piecewise-smooth loops at $x$ is path-connected. (No simple-connectedness assumption on $M$.) Fix $\gamma_0, \gamma_1 \in \Omega^0(x, x)$. By definition of null-homotopic, there exist continuous null-homotopies \begin{align*} H_0, H_1 : [0, 1]^2 \to M \end{align*} with $H_i(0, \cdot) = c_x$, $H_i(1, \cdot) = \gamma_i$, $H_i(s, 0) = H_i(s, 1) = x$ for all $s$ and $i = 0, 1$. By the smoothing theorem ([Whitney Embedding Theorem](/theorems/1519) applied to the manifold-with-corners $[0, 1]^2$ mapping into $M$, with smooth approximation rel boundary), each continuous null-homotopy can be replaced by a piecewise-smooth one homotopic rel boundary; the smoothed $H_i$ has the property that $s \mapsto H_i(s, \cdot)$ is a continuous path in the loop-space topology on $\Omega(x, x)$, valued in $\Omega^0(x, x)$ throughout (because each intermediate loop is null-homotopic by construction, namely null-homotopic via the restriction of $H_i$ to $[0, s] \times [0, 1]$). Define the continuous path \begin{align*} H : [0, 1] &\to \Omega^0(x, x), \\ H(s) &= \begin{cases} H_0(1 - 2s, \cdot) & s \in [0, 1/2] \\ H_1(2s - 1, \cdot) & s \in [1/2, 1] \end{cases}. \end{align*} Then $H(0) = H_0(1, \cdot) = \gamma_0$, $H(1/2) = H_0(0, \cdot) = c_x = H_1(0, \cdot)$, and $H(1) = H_1(1, \cdot) = \gamma_1$. Each intermediate value $H(s)$ is null-homotopic (exhibited by the partial homotopy of $H_0$ or $H_1$), so $H$ stays inside $\Omega^0(x, x)$. The two halves of $H$ glue continuously at $s = 1/2$ because both equal $c_x$ there. Hence $\Omega^0(x, x)$ is path-connected. [/step]

[step:Use continuity of $P$ to conclude path connectedness of $\mathrm{Hol}^0(M)$] The continuity of $P : \Omega(x, x) \to \mathrm{O}(n)$ was established in the proof of [Simple Connectivity and Path Connectivity](/theorems/2762): in any orthonormal frame on a precompact neighbourhood of the path's image, the parallel transport equation is a linear ODE \begin{align*} \dot V(t) = A_\gamma(t) V(t), \qquad V(0) = v, \end{align*} with $A_\gamma(t) \in \mathfrak{so}(n)$ continuous in $\gamma$ in our topology. The time-$1$ flow $P_\gamma = \Phi_\gamma(1)$ is given by the absolutely convergent Dyson series \begin{align*} \Phi_\gamma(1) = \sum_{k=0}^\infty \int_{0 \leq s_k \leq \cdots \leq s_1 \leq 1} A_\gamma(s_1) \cdots A_\gamma(s_k)\,d\mathcal{L}^k(s), \end{align*} which is a continuous functional of $A_\gamma \in C^0([0, 1]; \mathfrak{so}(n))$, hence continuous in $\gamma$. The continuity also restricts to the subspace $\Omega^0(x, x)$. Therefore $\mathrm{Hol}^0(M) = P(\Omega^0(x, x))$ is the continuous image of a path-connected space (path-connectedness of $\Omega^0(x, x)$ established above), hence is path-connected: for any $A_0, A_1 \in \mathrm{Hol}^0(M)$, choose $\gamma_0, \gamma_1 \in \Omega^0(x, x)$ with $P_{\gamma_i} = A_i$, and apply $P$ to a continuous path in $\Omega^0(x, x)$ from $\gamma_0$ to $\gamma_1$ produced by the same path-connectedness argument. The composition is a continuous path in $\mathrm{O}(n)$ from $A_0$ to $A_1$ lying entirely in $\mathrm{Hol}^0(M)$. [/step]

[step:Identify the path components of $\mathrm{O}(n)$ and locate the identity component] We use the topological structure of $\mathrm{O}(n)$. The orthogonal group $\mathrm{O}(n) \subset \mathbb{R}^{n \times n}$ is a smooth compact Lie group of dimension $n(n-1)/2$. The determinant homomorphism \begin{align*} \det : \mathrm{O}(n) &\to \{\pm 1\} \subseteq \mathbb{R}, \\ A &\mapsto \det A, \end{align*} is continuous (a polynomial in the matrix entries) and surjective. The preimages \begin{align*} \mathrm{SO}(n) &:= \det^{-1}(\{1\}) = \{A \in \mathrm{O}(n) : \det A = 1\}, \\ \mathrm{O}(n)^- &:= \det^{-1}(\{-1\}) = \{A \in \mathrm{O}(n) : \det A = -1\}, \end{align*} partition $\mathrm{O}(n)$ into two disjoint open subsets (open because $\det$ is continuous and $\{1\}, \{-1\}$ are open in $\{\pm 1\}$ with the discrete topology). They are also closed (each is the preimage of a closed singleton). So $\mathrm{O}(n)$ is the disjoint union of two clopen subsets, neither empty. It is a standard fact that $\mathrm{SO}(n)$ is path-connected (one can write any rotation as a product of two-plane rotations, each of which is connected to the identity by a $1$-parameter family of rotations). Concretely: any $A \in \mathrm{SO}(n)$ is conjugate (in $\mathrm{O}(n)$) to a block-diagonal matrix with $2 \times 2$ rotation blocks $R(\theta_i) = \begin{pmatrix} \cos\theta_i & -\sin\theta_i \\ \sin\theta_i & \cos\theta_i \end{pmatrix}$ and at most one $\pm 1$ scalar block (with even total number of $-1$'s, summing to $\det A = +1$). Each $R(\theta_i)$ is path-connected to the identity via $\theta_i \mapsto R(\theta_i)$ on $[0, \theta_i]$, so the product is path-connected to the identity. Path-connectedness in conjugates pulls back: $A$ is path-connected to $\mathrm{Id}$ in $\mathrm{SO}(n)$. Likewise $\mathrm{O}(n)^-$ is path-connected (any element is connected to a fixed reference reflection through analogous block-diagonal paths). Hence $\mathrm{O}(n)$ has exactly two path components: $\mathrm{SO}(n)$ (containing the identity $\mathrm{Id}$, which has determinant $+1$) and $\mathrm{O}(n)^-$ (containing reflections of determinant $-1$). The identity $\mathrm{Id} = P_{c_x} \in \mathrm{Hol}^0(M)$ because the constant loop $c_x$ is null-homotopic via the constant homotopy $H(s, t) = x$ for all $s, t$, and parallel transport along $c_x$ is the identity (the ODE has zero coefficients, so $V(t) \equiv v$). So $\mathrm{Hol}^0(M)$ contains $\mathrm{Id} \in \mathrm{SO}(n)$. [/step]

[step:Conclude that $\mathrm{Hol}^0(M) \subseteq \mathrm{SO}(n)$]We now combine the previous facts. $\mathrm{Hol}^0(M)$ is: \begin{enumerate} \item Path-connected (established via the continuity of $P$ above). \item A subset of $\mathrm{O}(n)$ (parallel transport is metric-preserving for the Levi-Civita connection). \item Contains the identity $\mathrm{Id}$ (since the constant loop is null-homotopic and parallel-transports as the identity). \end{enumerate} The path component of $\mathrm{O}(n)$ containing $\mathrm{Id}$ is $\mathrm{SO}(n)$ (by the determinant decomposition above). Since $\mathrm{Hol}^0(M)$ is path-connected and contains $\mathrm{Id}$, every element of $\mathrm{Hol}^0(M)$ is path-connected to $\mathrm{Id}$ within $\mathrm{Hol}^0(M)$, hence within $\mathrm{O}(n)$, hence lies in the path component of $\mathrm{Id}$: \begin{align*} \mathrm{Hol}^0(M) \subseteq \mathrm{SO}(n). \end{align*} This is the assertion of the theorem.[/step]

[guided]We have done all the heavy lifting in the preceding steps; the present step is the synthesis. The argument is a pure topological deduction with the structure: "a path-connected subset of a space, containing a point of a single path component, must lie entirely in that path component." All three ingredients required to invoke this principle have already been established, and we now collect them and apply it. **Why we are doing this synthesis at all.** The full holonomy group $\mathrm{Hol}(M) = P(\Omega(x, x))$ may include parallel transports along non-contractible loops, which can produce determinant-$(-1)$ elements (orientation-reversing isometries). The Klein bottle $K$ is the canonical witness: $K$ has a flat metric, so curvature-driven holonomy is the identity group $\{\mathrm{Id}\}$, yet the non-contractible loop produces a parallel transport with determinant $-1$, giving $\mathrm{Hol}(K) \cong \mathbb{Z}_2 \not\subseteq \mathrm{SO}(2)$. The restriction to null-homotopic loops kills this topological contribution, leaving only the curvature-driven part, which we will now show is necessarily orientation-preserving. **Ingredient 1: $\mathrm{Hol}^0(M)$ is path-connected.** This was established in the previous step. The argument was a two-stage application of "continuous image of path-connected is path-connected": first, $\Omega^0(x, x)$ is path-connected because every null-homotopic loop has a continuous null-homotopy to the constant loop (a topological fact independent of any assumption that $M$ be simply connected — we are restricted to the identity coset of $\Omega(x, x) / \Omega^0(x, x) = \pi_1(M, x)$). Second, $P : \Omega(x, x) \to \mathrm{O}(n)$ is continuous because the parallel transport ODE $\dot V = A_\gamma V$ has coefficients varying continuously with $\gamma$ and time-$1$ flow given by an absolutely convergent Dyson series. The continuous image $P(\Omega^0(x, x)) = \mathrm{Hol}^0(M)$ is therefore path-connected. **Ingredient 2: $\mathrm{Hol}^0(M) \subseteq \mathrm{O}(n)$.** This is the metric-preservation property of the Levi-Civita connection: for any loop $\gamma$, the parallel transport $P_\gamma$ is an isometry of $(T_x M, g_x)$, and after our chosen identification $T_x M \cong \mathbb{R}^n$ via an orthonormal basis, $P_\gamma \in \mathrm{O}(n)$. Since $\mathrm{Hol}^0(M)$ is the image of $\Omega^0(x, x) \subseteq \Omega(x, x)$ under $P$, and the image of $P$ lies in $\mathrm{O}(n)$, we have $\mathrm{Hol}^0(M) \subseteq \mathrm{O}(n)$. **Ingredient 3: $\mathrm{Id} \in \mathrm{Hol}^0(M)$.** The constant loop $c_x$ is null-homotopic via the constant homotopy $H(s, t) = x$, so $c_x \in \Omega^0(x, x)$. Parallel transport along $c_x$ has zero ODE coefficients, so $V(t) \equiv v$ and $P_{c_x} = \mathrm{Id}$. Therefore $\mathrm{Id} = P_{c_x} \in P(\Omega^0(x, x)) = \mathrm{Hol}^0(M)$. **The topological deduction.** We now combine the three ingredients. The path components of $\mathrm{O}(n)$ are exactly $\mathrm{SO}(n)$ (containing $\mathrm{Id}$) and $\mathrm{O}(n)^- = \mathrm{O}(n) \setminus \mathrm{SO}(n)$, distinguished by the continuous homomorphism $\det : \mathrm{O}(n) \to \{\pm 1\}$. Both $\mathrm{SO}(n)$ and $\mathrm{O}(n)^-$ are clopen in $\mathrm{O}(n)$ (preimages of $\{1\}$ and $\{-1\}$ under the continuous map $\det$, with $\{\pm 1\}$ carrying the discrete topology in which both singletons are clopen). Suppose, for contradiction, that some $A \in \mathrm{Hol}^0(M)$ has $\det A = -1$. Path-connectedness of $\mathrm{Hol}^0(M)$ produces a continuous path $\alpha : [0, 1] \to \mathrm{Hol}^0(M) \subseteq \mathrm{O}(n)$ with $\alpha(0) = \mathrm{Id}$ and $\alpha(1) = A$. The composition $\det \circ \alpha : [0, 1] \to \{\pm 1\}$ is continuous, with $(\det \circ \alpha)(0) = +1$ and $(\det \circ \alpha)(1) = -1$. But $[0, 1]$ is connected and $\{\pm 1\}$ is discrete, so any continuous map $[0, 1] \to \{\pm 1\}$ is constant — contradiction. Hence every $A \in \mathrm{Hol}^0(M)$ has $\det A = +1$, i.e., lies in $\mathrm{SO}(n)$. Equivalently and more abstractly: a path-connected subset of $\mathrm{O}(n)$ lies in a single path component, because path components partition $\mathrm{O}(n)$ into clopen pieces and a path-connected set cannot be split across two clopen pieces. The identity pins down the path component containing $\mathrm{Hol}^0(M)$ to be $\mathrm{SO}(n)$. We have shown: \begin{align*} \mathrm{Hol}^0(M) \subseteq \mathrm{SO}(n). \end{align*} **Why this fails for the full holonomy group.** $\mathrm{Hol}(M)$ in general is not path-connected: the identity component is $\mathrm{Hol}^0(M)$, and the quotient $\mathrm{Hol}(M) / \mathrm{Hol}^0(M)$ is the image of $\pi_1(M)$ under the holonomy map, which can have non-zero cosets. So $\mathrm{Hol}(M)$ may contain determinant-$(-1)$ elements coming from non-contractible loops, even though every element of $\mathrm{Hol}^0(M)$ is in $\mathrm{SO}(n)$ — the argument above breaks at Ingredient 1, since we can no longer connect a non-contractible loop's parallel transport to the identity by a path inside $\mathrm{Hol}(M)$. **Geometric meaning.** $\mathrm{Hol}^0(M) \subseteq \mathrm{SO}(n)$ says: parallel transport along null-homotopic loops always preserves orientation. This is consistent with the [Fundamental Principle of Riemannian Holonomy](/theorems/2764), which will identify orientability of $M$ with the existence of an $\mathrm{Hol}(M)$-invariant volume form: for orientable $M$, $\mathrm{Hol}(M) \subseteq \mathrm{SO}(n)$, and for general $M$, only the restricted part is orientation-preserving.[/guided]