Setin 2017
Setin 2017
Setin 2017
1.0
ED N =6
0.6
1 0.9 N =8
gc =
N = 10
0.5 0.8 N = 12
fimag
0.4 0.7
gc
0.3 0.6
0.2 0.5
1.0 1.5 2.0 2.5 3.0
tilt g
0.1
0.2 0.4 0.6 FIG. 5. Emergence of imaginary part of energy with
1 increasing tilt g. y axis is fraction of eigenvalues having
imaginary component for one realization of (1) at interaction
strength U = 0, disorder width W = 7, and system size N as
indicated. Dotted lines show our analytical results (28), (29)
FIG. 4. Critical tilt gc against inverse localization for the first few bifurcations.
length ⇠↵ 1 via exact diagonalization for 1000 disorder re-
alizations of (10) at system size N = 40 and lead parameters
as in Eq. (19). (We consider eigenstate ↵ = 20 of each real- so we can ignore the G[1:N ] term. If we assume
ization.) This confirms our analytical relationship (21).
1 1
h2|G[2:N ] |N i ⇠e (N 1)⇠
(25)
between any two sites, it is not obvious that the results of then
Sec. III C will carry over to ordinary chains with periodic
boundary conditions. Eq. 21, which connects gc and ⇠↵ g c ⇠ ⇠↵ 1 . (26)
for some eigenstate, requires a carefully fine-tuned lead
site. Can we do better? Can we take a generic lead
site—that is, a straightforward periodic chain? IV. MANY-PARTICLE CASE AND
If we are willing to relax our demands for rigor, we can INTERACTION BROADENING
make some estimates. Take the Hamiltonian (1) with pe-
riodic boundary conditions acting on one particle. Single A. Many-particle non-interacting case
out one arbitrary site for treatment as the “lead”, and
return to (12). Once again take E to be an eigenvalue of Now let the same Hamiltonian (1) act on many
the non-Hermitian Hamiltonian H(g), so (12) becomes particles—in fact on the half-filling sector—but take it
[1:N ] to be noninteractingQ(U = 0). Its eigenstates will be
0 = (E µ0 ) det(EI Hopen ) t20 det(EI [2:N ]
Hopen ) Slater determinants ↵2A c†↵ |0i with eigenvalues EA =
P
+ ( 1)N +1 t20 tN 1 (N +1)g
e . ↵2A E↵ . When two single-particle states pass through
(22) an exceptional point, developing imaginary parts to their
energies, they therefore take with them a whole class of
Take t0 = t—the supposed lead site is just a normal many-particle states.
lattice site, after all—and write the determinants in To quantify this e↵ect consider first increasing g
terms of the (1, N ) components of the Green’s functions through gc , the tilt at which the first two single-particle
[1:N ] [2:N ]
G[1:N ] , G[2:N ] of Hopen , Hopen . This becomes states go through an exceptional point. (In the example
above, of an open chain with a lead site, these will be
1 1 the lead site and the open-chain level n.) Call those two
0 = (E µ0 ) h1|G[1:N ] |N i + t h2|G[2:N ] |N i states ↵1 and ↵2 , and occupy a set A of additional lev-
(23)
+ e(N +1)g . els, not including ↵1 , ↵2 , with more particles. Since the
energy di↵erence of the many body state is the same as
Now work at gc . Once again write E↵ for the eigenvalue that of the delocalizing orbitals,
nearest µ0 ; even though t0 is no longer small, we expect
E↵1 A (g) E↵2 A (g) = E↵1 (g) E↵2 (g) (27)
1 1
E µ 0 ⇠ E↵ E ⇠ (E↵ µ) ⇠ p ⌧ 1 , (24)
2 L every such set gives a pair of levels that coalesce at gc .