Automated void identification by Blendmask: from hierarchical molecular gas to hierarchical voids in NGC 628

Abstract

We identify voids in NGC 628 from the JWST MIRI F770W image using a deep_learning method (BlendMask) and refine them by intensity contrast. These voids may be feedback_driven bubbles or dynamically formed structures. Cross_matching with archival star cluster/association catalogs shows that only up to 17.6% of voids are associated with such stellar populations. HST B_band peak_flux distributions of voids with and without these populations overlap substantially, suggesting many related clusters/associations remain unidentified or misclassified. Voids associated with star clusters/associations tend to have lower intensity contrast and larger sizes. An anti_correlation between void size and intensity contrast indicates larger voids have emptier centers, possibly due to stronger feedback. Thus, voids may provide a complementary tracer for identifying stellar populations and constraining their physical properties. To quantify spatial relationships among CO, 21$μ$m, H$_α$ sources, and voids, we construct networks linking each source pair. Among the nine networks, 21$μ$m and H$_α$ sources show the strongest spatial association. Compared to small voids, large voids exhibit progressively increasing separations from CO to 21$μ$m to H$_α$ sources to voids, consistent with an evolutionary sequence in space and time. Smaller voids lie closer to molecular clouds, while larger voids are more displaced. Compared with molecular clouds not associated with voids, those associated with voids are significantly more massive and appear more evolved. Indeed, 68% of molecular clouds associated with voids are also associated with 21$μ$m sources. These results support an evolutionary scenario where some voids originate within molecular clouds, grow through stellar feedback, and gradually detach from their parent clouds.