4

H then has matrix representation

EI H(g)
2 3
E µ0 t0
6 t0 E ✏1 t 7 (11) FIG. 3. Sketch of the lead spectrum as in Sec. III C:
6 t E ✏2 t 7
=6 7 E↵ is a level of the open chain, µ0 the chemical potential of
4 t E ✏3 t 5 the lead, and E the energy at which we measure the Green’s
t0 e(N +1)g t E ✏4 function. Unlabeled vertical lines are other eigenenergies of
the open chain.
and determinant
det(EI H(g)) = (E µ0 ) det(EI Hopen ) Hamiltonian with lead site. But we can choose these pa-
rameters to strongly constrain the non-Hermitian eigen-
t20 det(EI [2:N ]
Hopen ) (12)
value E, and hence relate gc , the tilt at which eigenstate
+( 1)N +2 t20 tN 1 (N +1)g
e ↵ coalesces with the lead state and develops an imaginary
part, to h1|↵ih↵|N i.
(expanding in minors along the first column). Since we Suppose we wish to probe the eigenstate ↵ of Hopen .
[1:N ]
take t0 small we can ignore the t20 term compared to the Then choose
t20 egN term. If we take E to be an eigenvalue of H(g)
this is 0 <E↵ µ0 ⌧ typical level spacing ,
1 (19)
[1:N ] t0 = (E↵ µ0 )
det(EI Hopen )
(13) 2
= ( 1)N tN 1
(E µ0 ) 1 2 (N +1)g
t0 e .
(cf Fig. 3). Because the open chain is localized, the
lead’s occupied state will not hybridize substantially with
any of the chain’s levels in the Hermitian chain. But as
B. Open-chain Green’s function
we increase g, the lead state and the chain level n will
start to hybridize, and the energy of the lead site and of
We can re-write the determinant in (13) in terms of state n will approach each other. When they coalesce,
the open-chain Green’s function. This has (1, N ) matrix which they will do at a value E ' 12 (µ0 + En ), both lev-
element els will develop imaginary parts. Because E↵ µ0 ⌧
[1:N ] [1:N ] 1 typical level spacing, we expect this to be the first pair
G1N (E) = [(EI Hopen ) ]1N
to coalesce. With
[1:N ] (14)
[adj (EI Hopen )]1N t20
= [1:N ] ⌘ '1 (20)
det(EI Hopen ) (E↵ E)(E µ0 )
for (where the estimate follows from our premeditated choice
[1:N ]
of t0 ), Eq. 18 will become
G1N (E) = ( 1)N +1 [det(EI [1:N ]
Hopen )] 1 N
t 1
. (15) h i
1
gc = ln h↵| c†1 cN |↵i
With this relation (13) becomes N +1
(21)
ln
(G1N )
(1:N ] 1
= (E µ0 ) 1 2 (N +1)g
t0 e . (16) = ⇠↵ 1 + ' ⇠↵ 1
N +1
We can extract the tunneling probability for an eigen- where we define an eigenstate localization length ⇠n 1 ⌘
[1:N ]
state |↵i of Hopen from the Green’s function ln h↵| c†1 cN |↵i.
We show gc and ⇠↵ for eigenstate ↵ = 20 of a chain
[1:N ]
h1|↵i h↵|N i = (E E↵ )G1N (E) (17) with N = 40 with 1000 disorder realizations in Fig. 4, and
see good agreement. We first diagonalize the open chain;
by identifying poles, so Eq. (16) is we then take µ0 = E20 0.01h/L and t = 0.005h/L,
in accordance with Eq. (19), and find gc in the result-
(N +1)g 1 1 2
e = h1|↵ih↵|N i(E↵ E) (E µ0 ) t0 . (18) ing closed chain. The variation comes from : E is not
always exactly E = 12 (E↵ + µ0 ).

C. Critical tilt gc
D. Closed chain without lead
Eq. 18 has three free parameters: µ0 , t0 , and g. E
is not a (continuously tunable) parameter: it is fixed by Even though the chain with lead has periodic boundary
µ0 , t0 , g, since it is an eigenvalue of the non-Hermitian conditions, in the sense that there are exactly two paths