how to approach this?

If all values are allowed to be the same, the k-smallest element can be anywhere: assuming values are unique.

In a max heap, a leaf contains the 1-smallest value, at index from N/arity upto N-1 or N depending on whether indexing starts at 0 or 1.
A leaf may contain any unique value up to the k = N - log_arity(N)th smallest (N-kth largest): large values will be closer to the root or, for k = N coincide with it.