The **strict inductive limit topology** is the canonical construction for building a locally convex topological vector space from a countable ascending chain of locally convex subspaces, each already equipped with its own topology. Rather than seeking a single family of seminorms controlling the entire space, the construction declares a set open if and only if its intersection with every piece of the chain is open in that piece. The resulting topology is the finest locally convex topology making each inclusion continuous, and it is characterised by a universal property that reduces [continuity](/page/Continuity) on the union to continuity on the pieces. The canonical application is the [test function space](/page/Test%20Function) $\mathcal{D}(\Omega)$, assembled from the Fréchet spaces $\mathcal{D}_K(\Omega)$ of smooth compactly supported [functions](/page/Function) — and the theory of [distributions](/page/Distribution) rests entirely on this inductive limit structure.

[motivation] ### The Inadequacy of a Single Fréchet Topology The most tractable locally convex spaces are Fréchet spaces: complete metrizable locally convex spaces whose topology is generated by a countable family of seminorms. The [Schwartz space](/page/Schwartz%20Space) $\mathcal{S}(\mathbb{R}^n)$ and $C^\infty(\Omega)$ are both Fréchet, and their topologies are fully described by explicit, manageable seminorm families. For many spaces that arise in PDE theory and functional analysis, however, no countable seminorm family suffices, because the objects one wishes to control are not uniformly bounded by any finite collection of constraints. The difficulty becomes concrete for $\mathcal{D}(\Omega) = \bigcup_{K \subset \Omega} \mathcal{D}_K(\Omega)$, where the union ranges over compact subsets $K \subset \Omega$. Each $\mathcal{D}_K(\Omega)$ — the space of smooth functions supported in the fixed compact set $K$ — is a [Fréchet space](/page/Fr%C3%A9chet%20Space) with seminorms \begin{align*} p_{K,N}(\varphi) &= \sum_{|\alpha| \leq N} \sup_{x \in K} |\partial^\alpha \varphi(x)|. \end{align*} Passing to the union destroys metrizability: any putative Fréchet topology on $\mathcal{D}(\Omega)$ would be generated by a countable family of seminorms $\{q_j\}$, and each $q_j$ can only control test functions whose supports are confined to some compact set $K_j$. Test functions whose supports escape every $K_j$ — that is, functions that probe the behaviour of the domain arbitrarily close to $\partial \Omega$ — lie beyond the reach of the entire family $\{q_j\}$, leaving the topology unable to detect convergence for such functions. ### The Inductive Approach The resolution is to derive the topology on the union directly from the topologies already present on the pieces, rather than imposing a new one from outside. The guiding constraint is that any reasonable topology on $\mathcal{D}(\Omega)$ must make each inclusion $\mathcal{D}_K(\Omega) \hookrightarrow \mathcal{D}(\Omega)$ continuous — if the topology on the union were coarser than that of some piece, one could not detect convergence on that piece, rendering the theory of distributions incomplete. Among all locally convex topologies satisfying this constraint, the correct choice is the *finest* one: the topology that declares as many [sets](/page/Set) open as possible, subject only to local convexity and continuity of the inclusions. This is the inductive limit topology. The universal property makes this choice canonical: a linear map out of the inductive limit is continuous if and only if its restriction to each piece is continuous. Every [topological](/page/Topology) question about $\mathcal{D}(\Omega)$ thereby reduces to a family of questions about its Fréchet pieces, where the standard tools of metrizable spaces apply. This reduction is not merely a convenience — it is the mechanism by which the entire distribution calculus becomes tractable. ### The Role of Strictness The word "strict" imposes a compatibility condition between the topology of each piece and the topology it inherits from the union. Without this condition, a sequence converging in $\mathcal{D}_K(\Omega)$ might fail to converge in $\mathcal{D}(\Omega)$ with supports confined to $K$ — an internal inconsistency that would make the relationship between pieces and the ambient space difficult to control. Strictness ensures that each piece sits inside the union as a topological subspace: the topology on $\mathcal{D}_K(\Omega)$ is exactly the subspace topology inherited from $\mathcal{D}(\Omega)$, so convergence in the piece is identical to convergence in the ambient space with all supports confined to that piece. Strictness is not a formal assumption but a geometric fact in the motivating example: $\mathcal{D}_{K_n}(\Omega)$ is a closed subspace of $\mathcal{D}_{K_{n+1}}(\Omega)$ because the support condition $\mathrm{supp}(\varphi) \subseteq K_n$ is preserved under uniform [limits](/page/Limit) of all derivatives — the support of a uniform limit of functions supported in $K_n$ is again contained in $K_n$. This closedness propagates through the chain and is precisely what makes the sequential characterisation of convergence in $\mathcal{D}(\Omega)$ so clean. [/motivation]

## Definition

The abstract framework applies to any countable chain of locally convex spaces. Throughout, all vector spaces are over $\mathbb{R}$; the complex case is identical.

The starting point is the algebraic structure: a nested family of spaces connected by injections that are compatible with the linear and topological structure.

[definition: Inductive System Of Locally Convex Spaces] An **inductive system of locally convex spaces** consists of a sequence $(V_n, \tau_n)_{n \in \mathbb{N}}$ of locally convex [topological vector spaces](/page/Topological%20Vector%20Space) together with, for each $n \in \mathbb{N}$, a continuous linear injection \begin{align*} \iota_n : V_n &\to V_{n+1}. \end{align*} For $m \leq n$, the **transition map** $\iota_{m,n} : V_m \to V_n$ is the composition \begin{align*} \iota_{m,n} &:= \iota_{n-1} \circ \cdots \circ \iota_m, \end{align*} with the convention $\iota_{n,n} := \mathrm{id}_{V_n}$. The transition maps satisfy the cocycle condition $\iota_{n,p} \circ \iota_{m,n} = \iota_{m,p}$ for all $m \leq n \leq p$. [/definition]

The algebraic inductive limit makes precise the sense in which the spaces $V_n$ are glued together along the transition maps.

[definition: Algebraic Inductive Limit] Let $(V_n, \iota_n)_{n \in \mathbb{N}}$ be an inductive system of locally convex spaces with transition maps $\iota_{m,n}$. The **algebraic inductive limit** is the quotient \begin{align*} V := \varinjlim_n V_n := \bigsqcup_{n=1}^\infty V_n \Big/ \sim, \end{align*} where the [equivalence relation](/page/Equivalence%20Relation) $\sim$ identifies $v \in V_m$ with $w \in V_n$ whenever there exists $N \geq m, n$ such that $\iota_{m,N}(v) = \iota_{n,N}(w)$. The **canonical injection** is the map \begin{align*} j_n : V_n &\to V \\ v &\mapsto [v]_\sim. \end{align*} [/definition]

The algebraic inductive limit carries a natural vector space structure. For $[v]_\sim, [w]_\sim \in V$ with representatives $v \in V_m$ and $w \in V_n$, choose any $N \geq m, n$ and define

\begin{align*} [v]_\sim + [w]_\sim &:= [\iota_{m,N}(v) + \iota_{n,N}(w)]_\sim, \\ \lambda [v]_\sim &:= [\lambda v]_\sim. \end{align*}

Well-definedness of addition: suppose $v \sim v'$ and $w \sim w'$ with $v \in V_m$, $v' \in V_{m'}$, $w \in V_n$, $w' \in V_{n'}$. Choose $N$ large enough that $\iota_{m,N}(v) = \iota_{m',N}(v')$ and $\iota_{n,N}(w) = \iota_{n',N}(w')$. Then $\iota_{m,N}(v) + \iota_{n,N}(w) = \iota_{m',N}(v') + \iota_{n',N}(w')$ in $V_N$, so the equivalence class is independent of the choice of representatives and of $N$ (replacing $N$ by any larger $N'$ and applying $\iota_{N,N'}$ preserves the equality by linearity). The vector space axioms — associativity, commutativity, existence of the zero element $[0_{V_1}]_\sim$, additive inverses $-[v]_\sim = [-v]_\sim$, and distributivity — are inherited from the $V_N$. Since each $\iota_n$ is injective, the canonical injections $j_n$ are injective, and $V = \bigcup_n j_n(V_n)$.

In the motivating example, $V_n = \mathcal{D}_{K_n}(\Omega)$ for an exhaustion $K_1 \subset K_2 \subset \cdots$ of $\Omega$ by compact sets with $K_j \subset \mathrm{int}(K_{j+1})$, and each $\iota_n$ is the natural inclusion (extending a function by zero outside its support). Since the inclusions are literal set-theoretic injections, $v \sim w$ if and only if $v = w$ as functions, and the quotient construction reduces to the ordinary union $\mathcal{D}(\Omega) = \bigcup_j \mathcal{D}_{K_j}(\Omega)$.

The algebraic inductive limit is a vector space without a topology. Any topology on $V$ that one proposes must make each canonical injection $j_n : V_n \hookrightarrow V$ continuous, since otherwise convergence in a piece would not imply convergence in the ambient space. The inductive limit topology is the finest such topology, so that the resulting dual space is as large as possible — ensuring that the maximum number of linear functionals are continuous.

[definition: Inductive Limit Topology] Let $(V_n, \iota_n)_{n \in \mathbb{N}}$ be an inductive system of locally convex spaces with algebraic inductive limit $V$ and canonical injections $j_n : V_n \to V$. The **inductive limit topology** on $V$ is the finest locally convex topology on $V$ such that $j_n$ is continuous for every $n \in \mathbb{N}$. [/definition]

The inductive limit topology admits a concrete description in terms of neighbourhoods: a convex balanced set $U \subseteq V$ is a neighbourhood of the origin in the inductive limit topology if and only if $j_n^{-1}(U)$ is a neighbourhood of the origin in $(V_n, \tau_n)$ for every $n \in \mathbb{N}$. This makes it explicit that the topology is generated by the family of all seminorms $p$ on $V$ whose restriction $p \circ j_n$ is continuous on $(V_n, \tau_n)$ for every $n$. The inductive limit topology is uniquely determined by the maximality condition together with local convexity.

However, there is a potential pathology: the subspace topology that the inductive limit induces on $V_n$ might be strictly coarser than $\tau_n$, meaning some [open sets](/page/Open%20Set) of $V_n$ fail to appear as intersections with open sets of $V$. The strict condition ensures this cannot happen.

[definition: Strict Inductive System] An inductive system $(V_n, \iota_n)_{n \in \mathbb{N}}$ is **strict** if for every $n \in \mathbb{N}$: 1. The image $\iota_n(V_n)$ is a closed subspace of $(V_{n+1}, \tau_{n+1})$. 2. The topology $\tau_n$ coincides with the subspace topology that $\iota_n(V_n)$ inherits from $(V_{n+1}, \tau_{n+1})$. [/definition]

[definition: Strict Inductive Limit] The **strict inductive limit** of a strict inductive system $(V_n, \iota_n)_{n \in \mathbb{N}}$ is the algebraic inductive limit $V = \varinjlim_n V_n$ equipped with the inductive limit topology. [/definition]

The closedness of $V_n$ in $V_{n+1}$ is the operative condition: it ensures that the embedding $j_n : V_n \to V$ is not merely a continuous injection but a topological embedding, so that the topology $\tau_n$ coincides with the subspace topology of $V_n$ in $V$. This is the content of the embedding theorem developed below.

[example: The Standard Strict Inductive System For D] Fix an open set $\Omega \subseteq \mathbb{R}^n$ and choose an exhaustion by compact sets $K_1 \subset K_2 \subset \cdots$ with $K_j \subset \mathrm{int}(K_{j+1})$ and $\bigcup_j K_j = \Omega$. For each $j \in \mathbb{N}$, the space $\mathcal{D}_{K_j}(\Omega) = \{\varphi \in C^\infty(\Omega) : \mathrm{supp}(\varphi) \subseteq K_j\}$ is a Fréchet space with the seminorms \begin{align*} p_{K_j, N}(\varphi) &= \sum_{|\alpha| \leq N} \sup_{x \in K_j} |\partial^\alpha \varphi(x)|, \quad N \in \mathbb{N}_0. \end{align*} Each inclusion $\iota_j : \mathcal{D}_{K_j}(\Omega) \hookrightarrow \mathcal{D}_{K_{j+1}}(\Omega)$ is continuous because $p_{K_{j+1}, N}(\varphi) \geq p_{K_j, N}(\varphi)$ for all $\varphi \in \mathcal{D}_{K_j}(\Omega)$ (the supremum is taken over a larger set but the function vanishes outside $K_j$, so the values coincide). The image $\iota_j(\mathcal{D}_{K_j}(\Omega))$ is closed in $\mathcal{D}_{K_{j+1}}(\Omega)$: if $\varphi_m \in \mathcal{D}_{K_j}(\Omega)$ and $\varphi_m \to \varphi$ in $\mathcal{D}_{K_{j+1}}(\Omega)$, then $\partial^\alpha \varphi_m \to \partial^\alpha \varphi$ uniformly on $K_{j+1}$ for every $\alpha$. For any $x \in K_{j+1} \setminus K_j$, each $\varphi_m(x) = 0$, so $\varphi(x) = 0$ by pointwise convergence. Hence $\mathrm{supp}(\varphi) \subseteq K_j$ and $\varphi \in \mathcal{D}_{K_j}(\Omega)$. Finally, the subspace topology on $\mathcal{D}_{K_j}(\Omega)$ from $\mathcal{D}_{K_{j+1}}(\Omega)$ coincides with $\tau_j$: both topologies are generated by seminorms of the form $\sum_{|\alpha| \leq N} \sup |\partial^\alpha \varphi|$, and for functions supported in $K_j$ the supremum over $K_{j+1}$ equals the supremum over $K_j$. The strict inductive limit $\mathcal{D}(\Omega) = \bigcup_j \mathcal{D}_{K_j}(\Omega)$ is independent of the choice of exhaustion — different exhaustions yield the same topology on $\mathcal{D}(\Omega)$. This independence follows from the universal property: if two exhaustions produce inductive limit topologies $\tau$ and $\tau'$, the identity map is continuous in both directions (since each $\mathcal{D}_K$ for any compact $K$ is contained in some piece of either exhaustion), forcing $\tau = \tau'$. [/example]