Hello, please ask a question about the depth of field of path traced rendering

I used pathtracing to render the EXR results in the render queue, and then used the depth of field channel in the resulting to blur in the post-production software, but I found that the boundaries between the foreground and the far ground were blurred inaccurately (even if I rendered the depth blur directly into the exr results), not as accurate as the results I did in MAYA or other DCC software.
Here’s a partial part of my rendering, it seems like the depth of field channel in the foreground is a few pixels larger than the RGB channel, and I don’t know if that’s causing my final blur to be inaccurate.
Is there any setting that can solve this problem, such as increasing sampling, or some console commands? Thank you very much, no one around me seems to know about this problem, and they don’t care too much. Ha ha