This paper asks how well a very simple boson-like picture of fermions near the Fermi surface captures the small extra energy called the correlation energy. The author shows that if one treats particle–hole excitations as a few completely collective bosonic degrees of freedom — in other words, as modes that are fully delocalized over the whole Fermi surface — the best such trial state gives about 92% of the correct leading-order correlation energy. The missing ~8% cannot be recovered without allowing more localized particle–hole structure.

The setup is a standard one in mathematical many-body physics: N spinless fermions in a three-dimensional periodic box, with interactions scaled in a mean-field way so that a small parameter ℏ scales like N−1/3. The Hartree–Fock approximation gives a baseline energy. The correlation energy is the difference between the true many-body ground-state energy and this Hartree–Fock energy and is a small negative correction of order ℏ. The paper builds trial states in the random phase approximation (RPA) framework, but restricts attention to trial states made from completely delocalized particle–hole pairs and optimizes over those states.

At a high level the work uses “approximate bosonization”: particle–hole pairs near the Fermi surface act approximately like bosons, and one can write quadratic (Bogoliubov-type) energies for these modes. There are two competing pictures in previous work: particle–hole pairs either delocalized over patches of the Fermi surface or sharply localized in momentum. Earlier rigorous results showed both pictures give the same leading-order correlation energy for nice interaction potentials. Here the author evaluates the extreme delocalized case and expands the energy to second order in the interaction. The simple collective model produces a second-order coefficient of about 0.28125, while the optimal RPA result proven earlier has coefficient 1 − log(2) ≈ 0.3068. The ratio is roughly 0.28125/0.3068 ≈ 0.92.