Abstract
In 1972, R.M. Erdahl investigated the convex structure of the set of $N$-representable 2-body reduced density matrices (RDMs) for fermions. While many of these results have become foundational in quantum chemistry and many-body physics, several aspects of the geometric structure remained unverified or specific to the fermionic case. This paper revisits Erdahl’s work, extending key results to the $m$-body setting and the more general quantum marginal problem. We identify a technical flaw in Erdahl’s proof regarding the relationship between extreme and exposed points in finite dimensions. Furthermore, we demonstrate that when $2m \ge N$, every extreme point of the set of $N$-representable $m$-body RDMs has a unique pre-image in both symmetric and anti-symmetric settings, a result that also extends to complementary quantum marginals.
Introduction
The $N$-representability problem is a cornerstone of quantum many-body theory. Because the complexity of an $N$-particle wavefunction grows exponentially with the number of particles, researchers typically focus on reduced density matrices (RDMs), which describe the behavior of small subsystems (usually $m=2$). However, not every $m$-body matrix corresponds to a valid $N$-body state. The set of all valid, $N$-representable RDMs forms a convex set, and understanding the geometry of this set—specifically its "corners" or extreme points—is vital for variational methods in quantum chemistry.
This work re-examines the classic results of R.M. Erdahl (1972) concerning the convex structure of these sets. We clarify the conditions under which a subsystem uniquely determines the whole and address a long-standing claim regarding the nature of the boundary of the $N$-representable set.
The Geometry of N-Representability
In the study of convex sets, two types of boundary points are of particular interest:
- Extreme Points: Points that cannot be expressed as a convex combination of other points in the set.
- Exposed Points: Points that are the unique minimizers of some linear functional (in physics, this usually corresponds to being the unique ground state of a specific Hamiltonian).
Erdahl originally claimed that in finite-dimensional spaces, every extreme point of the $N$-representable RDM set is also an exposed point. This is a significant claim because it implies that every "corner" of the state space corresponds to a unique physical ground state.
Identifying the Erdahl Flaw
Upon rigorous review, we found that Erdahl’s proof of the equivalence between extreme and exposed points (specifically Lemma 3.1) contains a subtle but critical error. The proof relied on the assumption that a specific combination of Hamiltonians would always result in a positive semi-definite operator. By employing a block matrix decomposition analysis, we show that this positivity is not guaranteed. While the claim that extreme points are exposed may still be true, it currently lacks a valid proof, leaving it as an open conjecture in the field.
Geometric Rigidity and Uniqueness
A primary contribution of this paper is the exploration of the "pre-image" of an RDM—the $N$-body state(s) that reduce to a given $m$-body RDM. We establish a threshold for geometric rigidity in these systems.
Theorem 2 formally proves that if the subsystem size $m$ is at least half the total system size ($2m \ge N$), then every extreme point of the $N$-representable $m$-body RDM set has a unique $N$-body pre-image. This holds for both bosons (symmetric) and fermions (anti-symmetric).
This result is extended to the Quantum Marginal Problem, which deals with general multi-partite systems without specific exchange symmetries. We show that for a pair of complementary marginals (an $m$-body part and an $(N-m)$-body part), if both are extreme points, the total $N$-body state is uniquely determined.
Interior Points and the Kernel of the Contraction Map
We also investigate the mapping (or contraction map) from the $N$-body space to the $m$-body space. If this mapping is not one-to-one—meaning the kernel of the map is non-trivial—the structure of the interior of the RDM set becomes interesting.
Theorem 8 shows that if the map has a kernel, then every point in the interior of the RDM set must have multiple $N$-body pre-images. Furthermore, at least two of these pre-images must lie on the boundary of the $N$-body state space. This highlights a sharp distinction between the unique determination of states at the "corners" (extreme points) and the ambiguity inherent in the interior of the set.
Discussion and Future Directions
While we have clarified the uniqueness of pre-images for $2m \ge N$, the case for smaller subsystems ($2m < N$) remains complex. In these regimes, results from quantum error-correcting codes (QECC) suggest that extreme RDMs can indeed have non-unique pre-images. Current evidence suggests that for fermions, the smallest system size where an extreme 2-body RDM might have a non-unique pre-image is $N=9$.
The question of whether every extreme point is exposed remains the most significant unresolved problem identified here. If an extreme point exists that is not exposed, it would imply the existence of "corners" in the RDM set that cannot be reached as the unique ground state of any linear Hamiltonian, posing a challenge to certain variational optimization strategies.
Summary updated