
3
(b) Atomic excitation spectrum
Frequency 𝜔−𝜇(eV)
Interaction 𝑈/𝑊
−8
−6
−4
−2
0
2
0.0 0.5 1.0 1.5 2.0 2.5
(a) Density of states 𝐴1(𝜔)at 𝑛=1.5
−2𝐽H
−(𝑈+𝐽H)
−2
0
2
4
0.5 1.0 1.5 2.0
𝑛=1
−8
−6
−4
−2
0
2
(c) →
Frequency 𝜔−𝜇(eV)
−8
−6
−4
−2
0
2
(d) →
Interaction 𝑈/𝑊
0.0 0.5 1.0 1.5 2.0 2.5−8
−6
−4
−2
0
2
(e) →
Figure 2. (a) Interaction Udependence of the itinerant orbital’s
density-of-states (DOS) A1(ω)obtained for the gKH model with
L= 48 sites, JH/U = 0.25, and n= 1.5. Results obtained with
DMRG using η= 0.04 broadening. The solid lines represent the
atomic-limit transitions. Inset depicts the results for half electron
filling n= 1. (b) Atomic excitation spectrum. For clarity, we mark
only the hole-like (electron removal) excitations and show only one
spin projection. D, T, S, H labels stand for doublon, triplet, singlet,
and holon, respectively. (c)-(e) DOS A1(ω)projected on the specific
final configurations: (c) parallel spins, (d) antiparallel spins, and (e)
on the holon; see the text for details. Results obtained with Lanczos
diagonalization of L= 8 lattice with broadening η= 0.05.
OSMP transition [9,10], the three-peak structure is visible in
A1(ω), and becomes clearer the larger the interaction Ube-
comes. Since the three-peak structure is most pronounced
for U≫W, it is instructive to examine the atomic limit
U, JH→ ∞ of the gKH model; see Fig. 2(b). The atom real-
izes the noninteger filling n= 1.5provided the ground states
(gs) of the 1- and 2-electron sectors are degenerate, which is
achieved at µ=U+JH/2. Then, the gs consists of a local in-
terorbital triplet, denoted as |T⟩, which is degenerate with an
itinerant doublon with localized spin, denoted as |D⟩. By re-
moving an electron from the triplet, one creates a holon in the
itinerant orbital (|T⟩→|H⟩), with the cost of energy U+JH.
Interestingly, from the doubly occupied state, one can remove
an electron in two different ways. Depending on the spin pro-
jection of the removed electron, one can arrive at a local triplet
or singlet, |D⟩→|T⟩or |D⟩→|S⟩, respectively. The former
is a zero-energy transition between degenerate states of the gs,
while the latter costs an energy 2JHas it breaks the Hund’s
rule. In Fig. 2(a), we plot the relevant energy scales of the
atomic limit (U+JHand 2JH) and find good agreement with
the full many-body calculations of the gKH chain.
Projections on the atomic configurations. To make a
stronger case for the atomic-limit interpretation of the three-
peak spectrum, we decompose the spectral function of the
full many-body calculation into individual transitions [29].
To this end, we use the projector Ponto specific config-
urations of the on-site Ising basis |γ= 1, γ = 2⟩, i.e.,
⟨⟨c†
γ,L/2,σ;Pcγ,L/2,σ⟩⟩h
ω[21]. For clarity, we discuss only
the hole part (below µ), as the electron part can be de-
scribed analogously. Upon removing an electron from the
itinerant orbital, we distinguish three contributions. (i) In
Fig. 2(c), we project onto the parallel-spin configuration,
P=|↑,↑⟩⟨↑,↑| +|↓,↓⟩⟨↓,↓|. The resulting weight forms
a band of excitations close to the Fermi level ω≃µ. This
transition is responsible for the metallic properties of the lat-
tice. (ii) In Fig. 2(d), we instead project onto the antiparal-
lel configuration, P=|↑,↓⟩⟨↑,↓| +|↓,↑⟩⟨↓,↑|. We observe
large weight in the middle band and some smaller weight at
ω≃µ. The middle band represents the interorbital singlet
which breaks the Hund’s rule: this is the 2JHHund excita-
tion. The band at ω≃µrepresents the Sz= 0 component
of the triplet (|↑,↓⟩ +|↓,↑⟩), costing zero energy to excite.
(iii) Finally, in Fig. 2(e), we project onto the holon configu-
ration, P=|0,↑⟩⟨0,↑| +|0,↓⟩⟨0,↓|. This gives the energet-
ically lowest band of excitations, which we recognize as the
LHB, arising from triplet to holon transitions. The starting
state needs to be a triplet because singlets are excluded from
the gs by the Hund’s rule.
Noninteger vs integer filling. As shown above, for non-
integer filling (doped system), the atomic limit is enough to
explain the Hund bands. When the atomic gs of adjacent
particle-number subspaces, say Nand N−1, are degenerate,
there is no cost Ufor the transition from the gs of subspace
Nto the gs of subspace N−1. The excitation cost is zero;
it is compensated by µwhich is tuned to cause the degener-
acy. However, if the N−1subspace contains not only the
gs but also higher multiplets, these multiplets can be accessed
in the photoemission process N→N−1with just the en-
ergy ∝JH. Analogous reasoning applies to inverse transitions
N−1→N. Thus, remarkably, this results in U-independent
Hund bands.
Consider now this behavior in a more general system, host-
ing more atomic configurations with different n. In Fig. 3we
present the three-orbital Hubbard model (3oH) results [60] for
various electron fillings. For n= 4.5, the atomic limit of our
setup [21] predicts one Hund excitation (between states with 5
and 4electrons) with energy 2JH[61], along with several U-
dependent Hubbard excitations. We pinpoint the Hund band
using the projector analysis, shown in Fig. 3(b). We differen-
tiate transitions arriving at |↑↓,↑,↑⟩ and |↑↓,↓,↑⟩. Similarly,
for the n= 3.5filling, the atomic limit implies Hund bands in
photoemission at 3JHand 5JH. They are shown in Fig. 3(c).
The 3JHband is a transition to a low-spin S= 1/2state [P
onto |↑,↓,↑⟩; see Fig. 3(a)]. The 5JHband originates in states