Figure 7

Single rising bubble

A rising bubble deforming from a sphere into a torus, with its frequency track and spectrograms

Figure 7 caption

Single rising bubble. A bubble deforms from a sphere into a torus as it rises (top, with frame times marked). We apply our model zero-shot to new bubbles outside the dataset, with no scene-specific fitting. BEM ground truth and the learned model both climb as the shape flattens while Minnaert stays fixed (bottom left; dashed lines mark the frames above). Our model stays within 1.65 % of BEM on average at a fraction of the cost, compared with Minnaert's 12.84 % error. The difference is audible in the rendered audio (bottom right).

From Table 1

For the single rising bubble, we solved BEM on all 282 mesh frames of the bubble in this figure, with no vertex cap; the other LBM scenes cap BEM at 5000 vertices.

Video

▶ Supplemental video — not yet added Drop the file in this folder and swap this block for the commented-out <video> element below.

Files

File	What it holds
`bub1_only_nocap_all_rows.csv`	one row per mesh frame: BEM frequency, assembly and solve times, surrogate frequency and timings, vertex counts
`bub1_bem_solve_timing.md`	the per-frame BEM ledger; the source of Table 1's `103603+2361` ms
`single_bubble_nn_timing.json`	the standalone surrogate pass; Table 1's `74.2` ms
`bub1_only_nocap_regressor_freqs.csv`	the baseline regressors on the same meshes
`vertex_scan.csv`	vertex count per mesh
`README.md`	Table 1's row for this scene, accounted cell by cell
`dataset/single_rising_bubble/`	the scene; its BEM column covers bubble 1 only, the bubble this figure follows