EmT: Locating empty territories of homology group generators in a dataset

Figure 1.  The three loops are equivalent in the sense that they all encircle the same hole. However, it can be desirable to find a representation that captures the empty territory, with no data point inside. The first two loops contain data points inside them, while the third one is empty. EmT is a method for finding the third type of representation at a specified level of complexity

Figure 2.  For a VR complex, if there are pairwise intersections between $ k $ points, the corresponding $ k $-simplex is added to the simplicial complex. (a) three circles with radius $ \frac{\delta}{2} $ that intersect pairwise, and $ a, b, c $ are within a distance of $ \delta $ from each other. The $ VR(X, \delta) $ includes three points $ a, b, c $, the edges $ \{ab, bc, ca\} $, and the triangle $ \{abc\} $. (b) the resulting $ VR(X, \delta) $

Figure 3.  (a) The same point cloud as in Fig. 1 is used to generate the persistence diagram of (b). The farther a point is from the diagonal, the longer it appears in the filtration. There are 40 $ H_0 $ generators and one prominent $ H_1 $ generator, which is consistent with the point cloud

Figure 4.  (a) The green circles occupy the area around the data points. The remaining area is the resulting alpha hull, outlined by the red curves. By straightening the arcs, it becomes an alpha shape, shown in blue lines. (b) A concave hull based on the alpha shape (blue polygon). (c) An inner shell (blue polygon) based on the same alpha shape as (a)

Figure 5.  (a), (b) and (c) are three different ways of occupying the area from the interior of the dataset with $ \alpha $ equal to 0.8. (d) An $ R_{\alpha, i} $ with $ \alpha $ equal to 0.6

Figure 6.  The same dataset from previous examples is used here to illustrate EmT. (a) Convex hull of the dataset in green lines. The red circles are set $ S_X $ and green crosses are set $ S_{X'} $. (b) The alpha hull in read lines and alpha shape in blue lines. Green circles are $ \alpha $ balls and green points are $ C_{\alpha} $. (c) Blue lines are $ CH_{\alpha} $ and red pluses are circle centers that are inside $ CH_{\alpha} $. (d) The comparison between the EmT output $ \widetilde{IS_{\alpha}} $ as blue diamonds and the original representation point set $ S_X $ returned by persistent homology as red circles

Figure 7.  The same outer loop is used in the four examples displayed in (a), (b), (c), and (d), but with different points inside. The black points are data points, the red circles are the raw representations from the persistent homology algorithm, the blue triangles are representations of the EmT algorithm, and in (c) the cyan curves are the corresponding alpha hull. In (a), the raw representation of persistent homology is not able to detect the inner points, while the EmT representation includes the inner points. In (b), both the raw representation and EmT have the inner points in their representations. In (c), while the raw representation contains the larger loop with the inner points, the EmT representation includes two separate loops, as indicated by the corresponding alpha hulls. In (d), due to the amount of scattered points inside the larger loop, the raw representation is not able to identify it; therefore the EmT representation is not able to find the larger loop either since EmT takes the raw representation as input

Figure 8.  A dataset from the homogeneous Poisson point process with intensity $ 300 $ in $ [0, 1]\times[0, 1] $. (a) Persistence diagram of the dataset. (b) The eight $ H_1 $ generators with the largest persistences reported by persistent homology in different colors. (c) The representations of the eight $ H_1 $ generators obtained by the EmT approach, where the boundary of the empty regions are better captured

Figure 9.  (a) A dataset generated from the uniform distribution in $ [0, 1]\times[0, 1]\times[0, 1] $; points inside the red regions are deleted from the dataset. (b) The persistence diagram of the dataset. (c) The raw representations from the persistent homology algorithm (blue and green points). The representations are displayed in blue and green shading, where the red points are observations inside the shaded volumes. (d) The representations from the EmT approach (blue and green points). The representations are displayed in blue and green shading, and there is no point inside the shading volumes

Figure 10.  (a) Cell tower locations in Minnesota where each of the black point indicates a cell tower. (b) The loop overlaps with the Red Lake Indian Reservation. (c) The persistence diagram of the dataset. (d), (e) Representations of eight $ H_1 $ generators with the largest persistences from persistent homology and the EmT approach, respectively

Figure 11.  (a) A Voronoi foam simulation. The eight large red points are the void seeds that generate eight void regions. The green crosses highlight one of the eight void regions. (b) Persistence diagram of the original dataset with 1200 points

Figure 12.  Persistence diagrams of the centers of k-means clustering for different k's. The k-means are made on the dataset shown in Fig. 11a

Figure 13.  (a) Cosmic voids representations ($ H_2 $ generators) reported by the persistent homology algorithm. (b) Cosmic voids representations from the EmT approach. As a qualitative comparison, the representations of EmT trace the empty volumes, while the raw representations reported by persistent homology do not define well the empty regions. (c) The raw representation of a particular cosmic void is displayed in the blue shading and the red points are galaxy inside the volume. (d) The EmT representation of a particular cosmic is displayed in the blue shading and there is no galaxy inside. The axis units are $ Mpc/h $, which megaparsec divided by a parameter $ h $ (which accounts for the uncertainty about the expansion rate of the universe). One parsec is approximately 3.26 lightyears