It is well-known that if $\mathrm{P} = \mathrm{NP}$, then the seemingly larger class $\mathrm{PH}$ also collapses to $\mathrm{P}$; hence, we may base the hardness of $\mathrm{NP}$ on that of $\mathrm{PH}$, despite the widespread belief that $\mathrm{PH} \not\subseteq \mathrm{NP}$. Although well established in the decision problem setting, few such “conditional collapse theorems” are known for $\textit{total}$ $\mathrm{NP}$ search problems. In this work, by building on the ideas of Ghentiyala and Li [GL25] and drawing an analogy with the polynomial hierarchy, we study conditional collapses in $\mathrm{TFNP}$. Our primary focus is on the $\mathrm{WeakPigeon}$ hierarchy: the class of problems obtained by $\textit{iterating}$ the $\mathrm{WeakPigeon}$ problem via oracles (e.g. $\mathrm{WeakPigeon}, \mathrm{WeakPigeon}^{\mathrm{WeakPigeon}}, \mathrm{WeakPigeon}^{(\mathrm{WeakPigeon}^{\mathrm{WeakPigeon}})}$, etc.).
First, we find that the problem of finding a multicollision of any fixed constant size (in a sufficiently compressing function) is contained in the $\mathrm{WeakPigeon}$ hierarchy, making it the first known $\mathrm{TFNP}$ problem where iteration non-trivially gives more power. Consequently, if $\mathrm{PWPP} = \mathrm{FP}$ then the problem of finding multicollisions also collapses to $\mathrm{FP}$. Our proof can be viewed as a reinterpretation of Buzek and Tessaro's "non-black-box" construction of collision resistant hash functions (CRHs) from multicollision-resistant hash functions (MCRHs) [BT24]. Interestingly, in the framework of $\mathrm{TFNP}$ problem iteration, this theorem follows from a black-box reduction, leading us to believe that $\mathrm{TFNP}$ problem iteration may be a useful explanatory framework for certain non-black-box constructions in cryptography. Our proposed shift in perspective also yields other new conditional collapses. For example, we see that if $\mathrm{PPP} = \mathrm{promiseUEOPL}$ or $\mathrm{PWPP} = \mathrm{LOSSY}$, then the problem of finding multicollisions also collapses to $\mathrm{promiseUEOPL}$ or $\mathrm{LOSSY}$ respectively.
Having demonstrated that the $\mathrm{WeakPigeon}$ hierarchy is surprisingly powerful, we next show that it nonetheless does not contain the $\mathrm{TFNP}$ class $\mathrm{PPAD}$ (and hence any of its superclasses e.g. $\mathrm{PPP}, \mathrm{PPA}$) under black-box reductions. Our separation utilizes the connection between $\mathrm{TFNP}$ classes and proof complexity: we prove that bounded-depth Frege has efficient proofs of the tautology asserting the totality of problems in the $\mathrm{WeakPigeon}$ hierarchy, while on the other hand the tautology asserting the totality of $\mathrm{PPAD}$ is known to be hard for bounded-depth Frege. These insights into the structure of iterated $\mathrm{TFNP}$ problems inform us about both the promise and limitations of using our techniques to obtain further conditional collapses in $\mathrm{TFNP}$.