Topological spaces and neighborhood filters

Jordan Bell

April 3, 2014

If XX is a set, a filter on XX is a set F\mathcal{F} of subsets of XX such that ∉F\emptyset \not \in \mathcal{F}; if A,BFA,B \in \mathcal{F} then ABFA \cap B \in \mathcal{F}; if AXA \subseteq X and there is some BFB \in \mathcal{F} such that BAB \subseteq A, then AFA \in \mathcal{F}. For example, if xXx \in X then the set of all subsets of XX that include xx is a filter on XX.1 A basis for the filter F\mathcal{F} is a subset BF\mathcal{B} \subseteq \mathcal{F} such that if AFA \in \mathcal{F} then there is some BBB \in \mathcal{B} such that BAB \subseteq A.

If XX is a set, a topology on XX is a set O\mathcal{O} of subsets of XX such that: ,XO\emptyset, X \in \mathcal{O}; if UαOU_\alpha \in \mathcal{O} for all αI\alpha \in I, then αIUαO\bigcup_{\alpha \in I} U_\alpha \in \mathcal{O}; if II is finite and UαOU_\alpha \in \mathcal{O} for all αI\alpha \in I, then αIUαO\bigcap_{\alpha \in I} U_\alpha \in \mathcal{O}. If NXN \subseteq X and xXx \in X, we say that NN is a neighborhood of xx if there is some UOU \in \mathcal{O} such that xUNx \in U \subseteq N. In particular, an open set is a neighborhood of every element of itself. A basis for a topology O\mathcal{O} is a subset B\mathcal{B} of O\mathcal{O} such that if xXx \in X then there is some BBB \in \mathcal{B} such that xBx \in B, and such that if B1,B2BB_1,B_2 \in \mathcal{B} and xB1B2x \in B_1 \cap B_2, then there is some B3BB_3 \in \mathcal{B} such that xB3B1B2x \in B_3 \subseteq B_1 \cap B_2.2

On the one hand, suppose that XX is a topological space with topology O\mathcal{O}. For each xXx \in X, let Fx\mathcal{F}_x be the set of neighborhoods of xx; we call Fx\mathcal{F}_x the neighborhood filter of xx. It is straightforward to verify that Fx\mathcal{F}_x is a filter for each xXx \in X. If NFxN \in \mathcal{F}_x, there is some UFxU \in \mathcal{F}_x that is open, and for each yUy \in U we have NFyN \in \mathcal{F}_y.

On the other hand, suppose XX is a set, for each xXx \in X there is some filter Fx\mathcal{F}_x, and: if NFxN \in \mathcal{F}_x then xNx \in N; if NFxN \in \mathcal{F}_x then there is some UFxU \in \mathcal{F}_x such that if yUy \in U then NFyN \in \mathcal{F}_y. We define O\mathcal{O} in the following way: The elements UU of O\mathcal{O} are those subsets of XX such that if xUx \in U then UFxU \in \mathcal{F}_x. Vacuously, O\emptyset \in \mathcal{O}, and it is immediate that XOX \in \mathcal{O}. If UαOU_\alpha \in \mathcal{O}, αI\alpha \in I and xU=αIUαx \in U=\bigcup_{\alpha \in I} U_\alpha then there is at least one αI\alpha \in I such that xUαx \in U_\alpha and so UαFxU_\alpha \in \mathcal{F}_x. As xUαUx \in U_\alpha \subseteq U and Fx\mathcal{F}_x is a filter, we get UFxU \in \mathcal{F}_x. If II is finite and UαIU_\alpha \in I, αI\alpha \in I, let U=αIUαU=\bigcap_{\alpha \in I} U_\alpha. If xUx \in U, then for each αI\alpha \in I, xUαx \in U_\alpha, and hence for each αI\alpha \in I, UαFxU_\alpha \in \mathcal{F}_x. As Fx\mathcal{F}_x is a filter, the intersection of any two elements of it is an element of it, and thus the intersection of finitely many elements of it is an element of it, so UFxU \in \mathcal{F}_x, showing that UOU \in \mathcal{O}. This shows that O\mathcal{O} is a topology. We will show that a set NN is a neighborhood of a point xx if and only if NFxN \in \mathcal{F}_x.

If NFxN \in \mathcal{F}_x, then let V={yN:NFy}V=\{y \in N: N \in \mathcal{F}_y\}. There is some U0FxU_0 \in \mathcal{F}_x such that if yU0y \in U_0 then NFyN \in \mathcal{F}_y. If yU0y \in U_0 then NFyN \in \mathcal{F}_y, which implies that yNy \in N, and hence U0VU_0 \subseteq V. U0VU_0 \subseteq V and U0FxU_0 \in \mathcal{F}_x imply that VFxV \in \mathcal{F}_x, which implies that xVx \in V. If yVy \in V then NFyN \in \mathcal{F}_y, and hence there is some UFyU \in \mathcal{F}_y such that if zUz \in U then NFzN \in \mathcal{F}_z. If zUz \in U then NFzN \in \mathcal{F}_z, which implies that zNz \in N, and hence UVU \subseteq V. UVU \subseteq V and UFyU \in \mathcal{F}_y imply that VFyV \in \mathcal{F}_y. Thus, if yVy \in V then VFyV \in \mathcal{F}_y, which means that VV is open, xVNx \in V \subseteq N tells us that NN is a neighborhood of xx.

If a set NN is a neighborhood of a point xx, then there is some open set UU with xUNx \in U \subseteq N. UU being open means that if yUy \in U then UFyU \in \mathcal{F}_y. As xUx \in U we get UFxU \in \mathcal{F}_x, and as UNU \subset N we get NFxN \in \mathcal{F}_x. Therefore a set NN is a neighborhood of a point xx if and only if NFxN \in \mathcal{F}_x.

In conclusion: If XX is a topological space and for each xXx \in X we define Fx\mathcal{F}_x to be the neighborhood filter of xx, then these filters satisfy the two conditions that if NFxN \in \mathcal{F}_x then xNx \in N and that if NFxN \in \mathcal{F}_x there is some UFxU \in \mathcal{F}_x such that if yUy \in U then NFyN \in \mathcal{F}_y. In the other direction, if XX is a set and for each point xXx \in X there is a filter Fx\mathcal{F}_x and the filters satisfy these two conditions, then there is a topology on XX such that these filters are precisely the neighborhood filters of each point.


  1. cf. François Trèves, Topological Vector Spaces, Distributions and Kernels, p. 6.↩︎

  2. cf. James R. Munkres, Topology, second ed., p. 78.↩︎