Localized computational issue after adaptive refinement in first octant

[GioPolCz]

Good evening,

I'm using Bitpit with octree grids for FV computations that involve adaptive refinement based on level set data. After multiple iterations of refinement, I’ve encountered a strange behavior, since the computed divergence of a scalar field shows consistent instabilities localized only in the first octant of the domain, while other octants remain unaffected.

To debug this, I created analytic test cases with known divergence, and I verified that the issue arises even before solving — when assigning the initial values of the starting field based on cell centroids. Specifically, the field appears corrupted or inconsistent in the first octant as soon as the refinement level increases beyond a certain point. However, this doesn’t happen for coarser grids (after 1–2 refinement steps the field is smooth and symmetric), and the behavior doesn’t lead to crashes or segmentation faults — only to clear numerical anomalies. This suggests that the issue may be related to how I manage data on the refined grid, and how data is associated with cells during or after refinement, rather than in the solver itself.

To better illustrate the issue, I’ve included a few snapshots showing the computed field for different refinement zones. The view is a cut plane of the 3D domain. The asymmetry becomes very clear there, and is always localized in the first octant, regardless of the coordinate of the geometry.

This for a refine with levelset localization of a sphere Here for a refinement box located in the center of the domain

Have you encountered something similar before, or do you have any idea what might cause a region-specific inconsistency like this?
Any insight would be greatly appreciated.

Best regards,
Giovanni