Non-Bloch Band Theory of Non-Hermitian Systems

Kazuki Yokomizo1 and Shuichi Murakami1, 2

1
Department of Physics, Tokyo Institute of Technology,
2-12-1 Ookayama, Meguro-ku, Tokyo, 152-8551, Japan
2
TIES, Tokyo Institute of Technology, 2-12-1 Ookayama, Meguro-ku, Tokyo, 152-8551, Japan
In spatially periodic Hermitian systems, such as electronic systems in crystals, the band structure
arXiv:1902.10958v2 [cond-mat.mes-hall] 12 Aug 2019

is described by the band theory in terms of the Bloch wave functions, which reproduce energy levels
for large systems with open boundaries. In this paper, we establish a generalized Bloch band theory
in one-dimensional spatially periodic tight-binding models. We show how to define the Brillouin
zone in non-Hermitian systems. From this Brillouin zone, one can calculate continuum bands, which
reproduce the band structure in an open chain. As an example, we apply our theory to the non-
Hermitian Su-Schrieffer-Heeger model. We also show the bulk-edge correspondence between the
winding number and existence of the topological edge states.

The band theory in crystals is fundamental for describ- |β1 | ≤ |β2 | ≤ · · · ≤ |β2M−1 | ≤ |β2M |, Cβ is given
ing electronic structure1 . By introducing the Bloch wave by the trajectory of βM and βM+1 under a condition
vector k, the band structure calculated within a unit cell |βM | = |βM+1 |. It is obtained as the condition to con-
reproduces that of a large crystal with open boundaries. struct continuum bands, which reproduce band structure
Here it is implicitly assumed that the electronic states are for a large crystal with open boundaries. We note that in
almost equivalent between a system with open bound- Hermitian systems, this condition reduces to Cβ : |β| = 1,
aries and one with periodic boundaries, represented by meaning that k becomes real. In previous works, systems
the Bloch wave function with real k. This is because the with M = 1 have been studied in general cases83 and in
electronic states extend over the system. limited cases95 .
Recently, non-Hermitian systems, which are described A byproduct of our theory is that one can prove the
by non-Hermitian Hamiltonians have been attracting bulk-edge correspondence. The bulk-edge correspon-
much attention. These systems have been both theo- dence has been discussed, but in most cases, it has not
retically and experimentally studied in many fields of been shown rigorously but by observation on some par-
physics2–71 . In particular, the bulk-edge correspondence ticular cases, together with an analogy to Hermitian sys-
has been intensively studied in topological systems. In tems. It in fact shows that the bulk-edge correspondence
contrast to Hermitian systems, it seems to be violated for the real Bloch wave vector cannot be true in non-
in some cases. The reasons for this violation have been Hermitian systems. In this paper, we show the bulk-
under debate52,72–94 . edge correspondence in the non-Hermitian Su-Schrieffer-
Heeger (SSH) model with the generalized Brillouin zone
One of the controversies is that in many previous and discuss the relationship between a topological invari-
works, the Bloch wave vector has been treated as real ant in the bulk and existence of the edge states.
in non-Hermitian systems, similarly to Hermitian ones. We start with a 1D tight-binding model, with its
In Ref. 83, it was proposed that in one-dimensional (1D) Hamiltonian given by
non-Hermitian systems, the wave number k becomes
N q
complex. The value of β ≡ eik is confined on a loop on X X X
the complex plane, and this loop is a generalization of the H= ti,µν c†n+i,µ cn,ν , (1)
Brillouin zone in Hermitian systems. In non-Hermitian n i=−N µ,ν=1

systems, the wave functions in large systems with open where N represents the range of the hopping and q repre-
boundaries do not necessarily extend over the bulk but sents the degrees of freedom per unit cell. This Hamilto-
are localized at either end of the chain, unlike those in nian can be non-Hermitian, meaning that ti,µν is not nec-
Hermitian systems. This phenomenon is called the non- essarily equal to t∗−i,νµ . Then one can write the real-space
Hermitian skin effect83 . Thus far, how to obtain the gen- eigen-equation as H |ψi = E |ψi, where the eigenvector is
eralized Brillouin zone has been known only for simple written as |ψi = (· · · , ψ1,1 , · · · , ψ1,q , ψ2,1 , · · · , ψ2,q , · · · )
T
systems. in an open chain. Thanks to the spatial periodicity, one
In this paper, we establish a generalized Bloch band can write the eigenvector as a linear combination:
theory in a 1D tight-binding model in order to deter- X n (j)
(j)
mine the generalized Brillouin zone Cβ for β ≡ eik , ψn,µ = φn,µ , φ(j)
n,µ = (βj ) φµ , (µ = 1, · · · , q). (2)
k ∈ C. First of all, we introduce the “Bloch” Hamil- j

tonian H (k) and rewrite it in terms of β as H (β). Then (j)

By imposing that φn,µ is an eigenstate, one can obtain
the eigenvalue equation det [H (β) − E] = 0 is an al-
the eigenvalue equation (for example, see Eq. (7)) for
gebraic equation for β, and let 2M be the degree of
β = βj as
the equation. The main result is that when the eigen-
value equation has solutions βi (i = 1, · · · , 2M ) with det [H (β) − E] = 0. (3)
 2

(a) Open chain (b) Generalized Bloch (a) t3 t3

Re E, t1+γ1/2 t1+γ1/2 t1+γ1/2

t2+γ2/2 t2+γ2/2
Im E
t1-γ1/2 t2-γ2/2 t1-γ1/2 t2-γ2/2 t1-γ1/2
A B A B A B
Continuum
(b) t2=1, t3=1/5, γ1=4/3, γ2=0 (c) t1=0.3, t2=1.1, t3=1/5, γ1=0 (d) t2=0.5, t3=1/5, γ1=5/3, γ2=1/3
bands 1
1 (b-1) t1=1.1 1 (c-1) γ2=4/3 (d-1) t1=0.3

Im(β)

Im(β)
Im(β)
L H(b), bÎCb

FIG. 1. Schematic figure of the band structure (a) in a finite -1 -1 -1

-1 Re(β) 1 -1 Re(β) 1 -1 Re(β) 1
open chain with various system sizes L, and (b) in the gen- (c-2) γ2= -4/3 (d-2) t1=-0.3
1
eralized Bloch Hamiltonian. The vertical axis represents the (b-2) t1=-1.1
distribution of the complex energy E. 1 1

Im(β)
Im(β)

Im(β)
-1
Here this eigenvalue equation is an algebraic equation for -1

β with an even degree 2M in general cases96 . -1 Re(β) 1 -1 Re(β) 1

-1
-1 Re(β) 1
One can see from Eq. (2) that β corresponds to the
Bloch wave number k ∈ R via β = eik in Hermitian sys- FIG. 2. (a) Non-Hermitian SSH model. The dotted boxes
tems. The bulk-band structure for reality of k reproduces indicate the unit cell. (b)-(d) Generalized Brillouin zone Cβ
the band structure of a long open chain. When extend- of this model. The values of the parameters are (b) t2 =
ing this idea to non-Hermitian systems, we should choose 1, t3 = 1/5, γ1 = 4/3, and γ2 = 0, with (b-1) t1 = 1.1 and (b-
values of β such that the bands of the Hamiltonian H (β) 2) t1 = −1.1; (c) t1 = 0.3, t2 = 1.1, t3 = 1/5, and γ1 = 0, with
reproduce those of a long open chain (Fig. 1). The levels (c-1) γ2 = 4/3 and (c-2) γ2 = −4/3; and (d) t2 = 0.5, t3 =
are discrete in a finite open chain, and as the system size 1/5, γ1 = 5/3, and γ2 = 1/3, with (d-1) t1 = 0.3 and (d-2)
becomes larger, the levels become dense and the asymp- t1 = −0.3.
totically form continuum bands (Fig. 1). Therefore, in or-
der to find the generalized Brillouin zone Cβ , one should
consider asymptotic behavior of level distributions in an
open chain in the limit of a large system size. In Her-
mitian systems, |β| is equal to unity, meaning that the
eigenstates extend over the bulk. On the other hand, in
non-Hermitian systems, |β| is not necessarily unity, and discussion in Secs. SII and SIII in the Supplemental Ma-
these states may be localized at either end of the chain. terial96 . We impose the wave function in Eq. (2) to rep-
Therefore, these bands cannot be called bulk bands, but resent an eigenstate. Apart from the positions near the
should be called continuum bands. These states are in- two ends, it leads to the eigenvalue equation (3). The
compatible with the periodic boundaries. The continuum boundary conditions place another constraint on the val-
bands are formed by changing β continuously along Cβ , ues of βi (i = 1, · · · , 2M ) in the form of an algebraic
as we show later. equation. We now suppose the system size L to be quite
Next. we find how to determine the generalized Bril- large and consider a condition to achieve densely dis-
louin zone Cβ , which determines the continuum bands. tributed levels (Fig. 1). The equation consists of terms of
Here we number the solutions βi (i = 1, · · · , 2M ) of L
the form (βi1 βi2 · · · βiM ) . When |βM | 6= |βM+1 |, there is
Eq. (3) so as to satisfy |β1 | ≤ |β2 | ≤ · · · ≤ |β2M−1 | ≤ L
only one leading term proportional to (βM+1 · · · β2M ) ,
|β2M |. We find that the condition to get the continuum
which does not allow continuum bands. Only when
bands can be written as
|βM | = |βM+1 |, are there two leading terms proportional
L L
|βM | = |βM+1 | , (4) to (βM βM+2 · · · β2M ) and to (βM+1 βM+2 · · · β2M ) . In
such a case, the relative phase between βM and βM+1
and the trajectory of βM and βM+1 gives Cβ . In Her- can be changed almost continuously for a large L, pro-
mitian systems, we can prove that Eq. (4) becomes ducing the continuum bands. We note that our condi-
|βM | = |βM+1 | = 196 , and Cβ is a unit circle, |β| = 1. tion Eq. (4) is independent of any boundary conditions.
When M = 1, this condition physically corresponds to In Ref. 83, it was proposed that the continuum bands
a condition for the formation of a standing wave in an require |βi | = |βj |. Nonetheless, this is not sufficient; ex-
open chain as proposed in Ref. 83. We discuss this point cept for the case |βM | = |βM+1 |, it does not allow the
in Sec. SI in the Supplemental Material96 . continuum bands.
To get Eq. (4), we focus on boundary conditions in an
open chain. Here we provide an outline of the process
by which we arrive at Eq. (4), and we give a detailed We apply Eq. (4) to the non-Hermitian SSH model as
 3

shown in Fig. 2(a). It is given by

X h γ1  †  γ1  †
H= t1 + cn,A cn,B + t1 − cn,B cn,A
n
2 2
 γ2  †  γ2  †
+ t2 + cn,B cn+1,A + t2 − cn+1,A cn,B
 2 i 2
+t3 c†n,A cn+1,B + c†n+1,B cn,A , (5)

where t1 , t2 , t3 , γ1 , and γ2 are real. The generalized Bloch

Hamiltonian H (β) can be obtained by a replacement
eik → β, similarly to Hermitian systems, as H (β) =
R+ (β) σ+ + R− (β) σ− , where σ± = (σx ± iσy ) /2, and
R± (β) are given by
 FIG. 3. (Color online) Phase diagram and bulk-edge corre-
γ2  −1  γ1  spondence with t3 = 1/5, γ1 = 5/3, and γ2 = 1/3. (a) Phase
R+ (β) = t2 − β + t1 + + t3 β,
2   2 diagram on the t1 -t2 plane. The blue region represents that
γ1 γ2  the winding number is 1, and the orange region represents
R− (β) = t3 β −1 + t1 − + t2 + β. (6)
2 2 that the system has exceptional points. Along the black ar-
row in (a) with t2 = 1.4, we show the results for (b) the
Therefore the eigenvalue equation can be written as winding number, (d) energy bands in a finite open chain, and
(e) the continuum bands from the generalized Brillouin zone
R+ (β) R− (β) = E 2 , (7) Cβ . The edge states are shown in red in (d). (c) shows ℓ+
(red) and ℓ− (blue) on the R plane with t1 = 1 and t2 = 1.4.
which is a quartic equation for β; i.e., M = 2, having
four solutions βi (i = 1, · · · , 4) satisfying |β1 | ≤ |β2 | ≤
|β3 | ≤ |β4 |. Then Eq. (4) is given by |β2 | = |β3 |96 .
as we change one parameter, the gap closes at E = 0, and
The trajectory of β2 and β3 satisfying the condition
w+ and −w− change by 1 at the same time, giving rise to
|β2 | = |β3 | determines the generalized Brillouin zone Cβ ,
the change of the winding number by unity. On the other
and it is shown in Figs. 2(b)-2(d) for various values of the
hand, in case (b), only one of the two coefficients R± (β)
parameters. It always forms a loop enclosing the origin
becomes zero, and it represents an exceptional point.
on the complex plane. Nonetheless, we do not have a rig-
We obtain the phase diagram on the t1 -t2 plane in
orous proof that Cβ is always a single loop encircling the
origin. We find some features of Cβ . First, our result does Fig. 3(a) and on the γ1 -γ2 plane in Fig. 4(a). In these
phase diagrams, the winding number w is 1 in the blue
not depend on whether |β| is larger or smaller than unity,
region. By definition, w changes only when R± (β) = 0
as opposed to the suggestions in previous works57,83 ; in
on the generalized Brillouin zone Cβ , and the gap closes.
Fig. 2(d-2), |β| takes both values more than 1 and values
The energy bands in a finite open chain calculated along
less than 1. Second, Cβ can be a unit circle even for non-
the black arrow in Fig. 3(a) are shown in Fig. 3(d), and
Hermitian cases; for example, when t1 = t3 = γ2 = 0.
one can confirm that the edge states appear in the region
Finally, Cβ can have cusps, corresponding to the cases
where w = 1. In addition, the continuum bands using Cβ
where three solutions share the same absolute value96 .
(Fig. 3 (e)) agree with these energy bands. In Fig. 4(b),
We calculate the winding number w for the Hamilto-
we give the energy bands calculated along the green ar-
nian H(β). Thanks to the chiral symmetry, w can be
row in Fig. 4(a), and the edge states appear similarly to
defined as96
Fig. 3(d). On the other hand, the system has the excep-
w+ − w− 1 tional points in the orange region. The phase with the
w=− , w± = [arg R± (β)]Cβ , (8)
2 2π exceptional points extends over a finite region96 .
We discuss the bulk-edge correspondence in our model.
where [arg R± (β)]Cβ means the change of the phase of The loops ℓ± drawn by R± (β) on the R plane are shown
R± (β) as β goes along the generalized Brillouin zone in Fig. 3(c) and in Figs. 4(c) and 4(d) for certain values
Cβ in a counterclockwise way. It is proposed that w of the parameters. In both Fig. 3(c) and Fig. 4(c), the
corresponds to the presence or absence of the topological system has the winding number w = 1, since both ℓ+ and
edge states83 . ℓ− surround the origin O, leading to w+ = −1 and w− =
We show how the gap closes in our model. It closes 1. In Fig. 4(a), one can continuously change the values of
when E = 0, i.e., R+ (β) = 0 or R− (β) = 0. Let the parameters to the Hermitian limit, γ1 , γ2 → 0, while
β = βia (i = 1, 2, a = +, −) denote the solutions of the keeping the gap open and while w = 1 remains. The
equation Ra (β) = 0, with |β1a | ≤ |β2a |. When E = 0 is in same is true for Fig. 3(a). Therefore, by following the
the continuum bands, Eq. (4) should be satisfied for the proof in Hermitian cases97 , one can prove the bulk-edge
four solutions βi± (i = 1, 2). It can be classified into two correspondence even for the non-Hermitian cases, and
cases, (a) |β1a | ≤ |β2a | = β1−a ≤ β2−a (a = +, −), and the existence of zero-energy states is derived96 . On the
(b) |β1a | ≤ β1−a = β2−a ≤ |β2a | (a = +, −). In case (a), other hand, ℓ− passes O as shown in Fig. 4(d), where the
 4

β → 1/β. Suppose the numbering of the sites is reversed

by setting n′ = L + 1 − n for the site index n(= 1, · · · , L);
then β becomes β ′ = 1/β, but the form of the condition
′ ′
is invariant: |βM | = βM+1 .
Through this definition of the continuum bands, one
can show the bulk-edge correspondence without ambigu-
ity by defining the winding number w from the general-
ized Brillouin zone in 1D systems with chiral symmetry.
Indeed, we showed that the zero-energy states appear in
the non-Hermitian SSH model when w takes nonzero val-
ues, and we also revealed that these states correspond to
topological edge states. It is left for future works to deter-
mine how to calculate the continuum bands for systems
with other symmetries.
FIG. 4. Phase diagram and bulk-edge correspondence with The construction of the generalized Brillouin zone
t1 = 0, t2 = 1, and t3 = 1/5. (a) Phase diagram on the γ1 -γ2 can be extended to higher dimensions as well. In two-
plane. The blue region represents that the winding number dimensional (2D)
systems, we introduce

the two param-
is 1, and the orange region represents that the system has eters β x = eikx and β y = eiky . Then the eigenvalue
exceptional points. (b) Energy bands calculated along the equation det [H (β x , β y ) − E] = 0, where H (β x , β y ) is a
green arrow in (a) with γ2 = 1.4. Note that γc ≃ 1.89. The 2D generalized Bloch Hamiltonian, is an algebraic equa-
edge states are shown in red. (c),(d) Loops ℓ+ (red) and tion for β x and β y . If we fix β y (β x ), this system can be
ℓ− (blue) on the R plane. The values of the parameters are regarded as a 1Dsystem, and the criterion is given by
(c) γ1 = −1 and γ2 = 1.4, and (d) γ1 = 2.1 and γ2 = 1.4. x x y y
Note that ℓ− passes the origin in (d), which corresponds to βM x
= βM x +1
βM y
= βM y +1
, where 2Mx (2My )
exceptional points. is the degree of the eigenvalue equation for β x (β y ). Thus,
we can get the conditions for the continuum bands. Nev-
ertheless, it is still an open question how to determine
system has exceptional points. We note that the winding the generalized Brillouin zone in higher dimensions.
number is not well defined in this case.
In summary, we establish a generalized Bloch band the- We also apply our theory to the tight-binding model
ory in 1D tight-binding systems and obtain the condition proposed in Ref. 74, and we show that the Bloch wave
for the continuum bands. We show the way to construct number k has a nonzero imaginary part, and the bulk-
the generalized Brillouin zone Cβ , which is fundamen- edge correspondence can be established with k ∈ C96 .
tal for obtaining the continuum bands. Here the Bloch We conclude that some previous works on the bulk-edge
wave number k takes complex values in non-Hermitian correspondence using the reality of the Bloch wave vector
systems. Our conclusion, |βM | = |βM+1 |, is physically require further investigation.
reasonable in several aspects. First, it is independent of This work was supported by a Grant-in-Aid for
any boundary conditions. Thus, for a long open chain, Scientific Research (Grants No. JP18H03678 and
irrespective of any boundary conditions, the spectrum No. JP16J07354) by MEXT, Japan; by CREST, JST
asymptotically approaches the same continuum bands (No. JP-MJCR14F1); and by the MEXT Elements Strat-
calculated from Cβ 96 . Second, it reproduces the known egy Initiative to Form Core Research Center (TIES).
result in the Hermitian limit, i.e., |β| = 1. Third, the K. Y. was also supported by JSPS KAKENHI (Grant
form of the condition is invariant under the replacement No. 18J22113).

1 8
C. Kittel, P. McEuen, and P. McEuen, Introduction to J. Carlström and E. J. Bergholtz,
Solid State Physics, Vol. 8 (Wiley, New York, 1996). Phys. Rev. A 98, 042114 (2018).
2 9
J. González and R. A. Molina, T. M. Philip, M. R. Hirsbrunner, and M. J. Gilbert,
Phys. Rev. B 96, 045437 (2017). Phys. Rev. B 98, 155430 (2018).
3 10
V. Kozii and L. Fu, arXiv:1708.05841. Y. Chen and H. Zhai, Phys. Rev. B 98, 245130 (2018).
4 11
A. A. Zyuzin and A. Y. Zyuzin, K. Moors, A. A. Zyuzin, A. Y. Zyuzin, R. P. Tiwari, and
Phys. Rev. B 97, 041203(R) (2018). T. L. Schmidt, Phys. Rev. B 99, 041116(R) (2019).
5 12
H. Shen and L. Fu, Phys. Rev. Lett. 121, 026403 (2018). R. Okugawa and T. Yokoyama,
6
R. A. Molina and J. González, Phys. Rev. B 99, 041202(R) (2019).
13
Phys. Rev. Lett. 120, 146601 (2018). J. C. Budich, J. Carlström, F. K. Kunst, and E. J.
7
T. Yoshida, R. Peters, and N. Kawakami, Bergholtz, Phys. Rev. B 99, 041406(R) (2019).
14
Phys. Rev. B 98, 035141 (2018). Z. Yang and J. Hu, Phys. Rev. B 99, 081102(R) (2019).
 5

15 40
T. Yoshida, R. Peters, N. Kawakami, and Y. Hatsugai, M. A. Bandres, S. Wittek, G. Harari, M. Parto, J. Ren,
Phys. Rev. B 99, 121101(R) (2019). M. Segev, D. N. Christodoulides, and M. Khajavikhan,
16
Y. Wu, W. Liu, J. Geng, X. Song, X. Ye, C.-K. Duan, Science 359, eaar4005 (2018).
41
X. Rong, and J. Du, Science 364, 878 (2019). M. Pan, H. Zhao, P. Miao, S. Longhi, and L. Feng, Nat.
17
P. San-Jose, J. Cayao, E. Prada, and R. Aguado, Sci. Rep. Commun. 9, 1308 (2018).
42
6, 21427 (2016). L. Jin and Z. Song, Phys. Rev. Lett. 121, 073901 (2018).
18 43
Q.-B. Zeng, B. Zhu, S. Chen, L. You, and R. Lü, S. Malzard and H. Schomerus,
Phys. Rev. A 94, 022119 (2016). Phys. Rev. A 98, 033807 (2018).
19 44
C. Li, X. Z. Zhang, G. Zhang, and Z. Song, Z. Oztas and C. Yuce, Phys. Rev. A 98, 042104 (2018).
45
Phys. Rev. B 97, 115436 (2018). M. Kremer, T. Biesenthal, L. J. Maczewsky, M. Heinrich,
20
K. Kawabata, Y. Ashida, H. Katsura, and M. Ueda, R. Thomale, and A. Szameit, Nat. Commun. 10, 435
Phys. Rev. B 98, 085116 (2018). (2019).
21 46
A. Guo, G. J. Salamo, D. Duchesne, R. Morandotti, K. Y. Bliokh, D. Leykam, M. Lein, and F. Nori, Nat.
M. Volatier-Ravat, V. Aimez, G. A. Siviloglou, and D. N. Commun. 10, 580 (2019).
47
Christodoulides, Phys. Rev. Lett. 103, 093902 (2009). S. Wang, B. Hou, W. Lu, Y. Chen, Z. Zhang, and C. Chan,
22
C. E. Rüter, K. G. Makris, R. El-Ganainy, D. N. Nat. Commun. 10, 832 (2019).
48
Christodoulides, M. Segev, and D. Kip, Nat. Phys. 6, 192 S. Chen, W. Zhang, B. Yang, T. Wu, and X. Zhang, Sci.
(2010). Rep. 9, 5551 (2019).
23 49
L. Feng, M. Ayache, J. Huang, Y.-L. Xu, M.-H. Lu, Y.- T. E. Lee and C.-K. Chan, Phys. Rev. X 4, 041001 (2014).
50
F. Chen, Y. Fainman, and A. Scherer, Science 333, 729 Y. Xu, S.-T. Wang, and L.-M. Duan,
(2011). Phys. Rev. Lett. 118, 045701 (2017).
24 51
A. Regensburger, C. Bersch, M.-A. Miri, G. Onishchukov, Y. Ashida, S. Furukawa, and M. Ueda, Nat. Commun. 8,
D. N. Christodoulides, and U. Peschel, Nature (London) 15791 (2017).
52
488, 167 (2012). Z. Gong, Y. Ashida, K. Kawabata, K. Takasan, S. Hi-
25
L. Feng, Y.-L. Xu, W. S. Fegadolli, M.-H. Lu, J. E. gashikawa, and M. Ueda, Phys. Rev. X 8, 031079 (2018).
53
Oliveira, V. R. Almeida, Y.-F. Chen, and A. Scherer, Nat. M. Nakagawa, N. Kawakami, and M. Ueda,
Mater. 12, 108 (2013). Phys. Rev. Lett. 121, 203001 (2018).
26 54
C. Poli, M. Bellec, U. Kuhl, F. Mortessagne, and K. Takata and M. Notomi,
H. Schomerus, Nat. Commun. 6, 6710 (2015). Phys. Rev. Lett. 121, 213902 (2018).
27 55
B. Zhen, C. W. Hsu, Y. Igarashi, L. Lu, I. Kaminer, L. Pan, S. Chen, and X. Cui,
A. Pick, S.-L. Chua, J. D. Joannopoulos, and M. Soljačić, Phys. Rev. A 99, 011601(R) (2019).
56
Nature (London) 525, 354 (2015). J. Li, A. K. Harter, J. Liu, L. de Melo, Y. N. Joglekar,
28
H. Zhao, S. Longhi, and L. Feng, Sci. Rep. 5, 17022 (2015). and L. Luo, Nat. Commun. 10, 855 (2019).
29 57
K. Ding, Z. Q. Zhang, and C. T. Chan, T. Liu, Y.-R. Zhang, Q. Ai, Z. Gong,
Phys. Rev. B 92, 235310 (2015). K. Kawabata, M. Ueda, and F. Nori,
30
S. Weimann, M. Kremer, Y. Plotnik, Y. Lumer, S. Nolte, Phys. Rev. Lett. 122, 076801 (2019).
58
K. Makris, M. Segev, M. Rechtsman, and A. Szameit, Nat. M. S. Rudner and L. S. Levitov,
Mater. 16, 433 (2017). Phys. Rev. Lett. 102, 065703 (2009).
31 59
H. Hodaei, A. U. Hassan, S. Wittek, H. Garcia-Gracia, J. M. Zeuner, M. C. Rechtsman, Y. Plotnik, Y. Lumer,
R. El-Ganainy, D. N. Christodoulides, and M. Kha- S. Nolte, M. S. Rudner, M. Segev, and A. Szameit,
javikhan, Nature (London) 548, 187 (2017). Phys. Rev. Lett. 115, 040402 (2015).
32 60
W. Chen, Ş. K. Özdemir, G. Zhao, J. Wiersig, and K. Mochizuki, D. Kim, and H. Obuse,
L. Yang, Nature (London) 548, 192 (2017). Phys. Rev. A 93, 062116 (2016).
33 61
P. St-Jean, V. Goblot, E. Galopin, A. Lemaı̂tre, T. Ozawa, L. Xiao, X. Zhan, Z. Bian, K. Wang, X. Zhang, X. Wang,
L. Le Gratiet, I. Sagnes, J. Bloch, and A. Amo, Nat. J. Li, K. Mochizuki, D. Kim, N. Kawakami, et al., Nat.
Photonics 11, 651 (2017). Phys. 13, 1117 (2017).
34 62
B. Bahari, A. Ndao, F. Vallini, A. El Amili, Y. Fainman, N. Hatano and D. R. Nelson,
and B. Kanté, Science 358, 636 (2017). Phys. Rev. Lett. 77, 570 (1996).
35 63
J. Wang, H. Y. Dong, Q. Y. Shi, W. Wang, and K. H. N. Hatano and D. R. Nelson,
Fung, Phys. Rev. B 97, 014428 (2018). Phys. Rev. B 56, 8651 (1997).
36 64
H. Zhou, C. Peng, Y. Yoon, C. W. Hsu, K. A. Nelson, N. Hatano and D. R. Nelson,
L. Fu, J. D. Joannopoulos, M. Soljačić, and B. Zhen, Phys. Rev. B 58, 8384 (1998).
65
Science 359, 1009 (2018). J. A. S. Lourenço, R. L. Eneias, and R. G. Pereira,
37 Phys. Rev. B 98, 085126 (2018).
M. Parto, S. Wittek, H. Hodaei, G. Harari, M. A.
66
Bandres, J. Ren, M. C. Rechtsman, M. Segev, E. I. Rosenthal, N. K. Ehrlich, M. S. Rud-
D. N. Christodoulides, and M. Khajavikhan, ner, A. P. Higginbotham, and K. W. Lehnert,
Phys. Rev. Lett. 120, 113901 (2018). Phys. Rev. B 97, 220301(R) (2018).
38 67
H. Zhao, P. Miao, M. H. Teimourpour, S. Malzard, R. El- K. Luo, J. Feng, Y. Zhao, and R. Yu, arXiv:1810.09231.
68
Ganainy, H. Schomerus, and L. Feng, Nat. Commun. 9, M. Wang, L. Ye, J. Christensen, and Z. Liu,
981 (2018). Phys. Rev. Lett. 120, 246601 (2018).
39 69
G. Harari, M. A. Bandres, Y. Lumer, M. C. Rechtsman, M. Ezawa, Phys. Rev. B 99, 121411(R) (2019).
70
Y. D. Chong, M. Khajavikhan, D. N. Christodoulides, and M. Ezawa, Phys. Rev. B 99, 201411(R) (2019).
71
M. Segev, Science 359, eaar4003 (2018). M. Ezawa, Phys. Rev. B 100, 045407 (2019).
72
Y. C. Hu and T. L. Hughes,
 6

88
Phys. Rev. B 84, 153101 (2011). L. Jin and Z. Song, Phys. Rev. B 99, 081103(R) (2019).
73 89
K. Esaki, M. Sato, K. Hasebe, and M. Kohmoto, H. Wang, J. Ruan, and H. Zhang,
Phys. Rev. B 84, 205128 (2011). Phys. Rev. B 99, 075130 (2019).
74 90
T. E. Lee, Phys. Rev. Lett. 116, 133903 (2016). D. S. Borgnia, A. J. Kruchkov, and R.-J. Slager,
75
D. Leykam, K. Y. Bliokh, C. Huang, Y. D. Chong, and arXiv:1902.07217.
91
F. Nori, Phys. Rev. Lett. 118, 040401 (2017). Z. Ozcakmakli Turker and C. Yuce,
76
V. M. Martinez Alvarez, J. E. Barrios Vargas, and L. E. F. Phys. Rev. A 99, 022127 (2019).
92
Foa Torres, Phys. Rev. B 97, 121401(R) (2018). E. Edvardsson, F. K. Kunst, and E. J. Bergholtz,
77
Y. Xiong, J. Phys. Commun. 2, 035043 (2018). Phys. Rev. B 99, 081302(R) (2019).
78 93
H. Shen, B. Zhen, and L. Fu, C.-H. Liu, H. Jiang, and S. Chen,
Phys. Rev. Lett. 120, 146402 (2018). Phys. Rev. B 99, 125103 (2019).
79 94
C. Yuce, Phys. Rev. A 97, 042118 (2018). C. H. Lee and R. Thomale,
80
C. Yin, H. Jiang, L. Li, R. Lü, and S. Chen, Phys. Rev. B 99, 201103(R) (2019).
95
Phys. Rev. A 97, 052115 (2018). F. K. Kunst and V. Dwivedi,
81
C. Yuce, Phys. Rev. A 98, 012111 (2018). Phys. Rev. B 99, 245116 (2019).
82 96
F. K. Kunst, E. Edvardsson, J. C. Budich, and E. J. See Supplemental Material for general description of the
Bergholtz, Phys. Rev. Lett. 121, 026808 (2018). non-Bloch band theory, derivation of the condition for con-
83
S. Yao and Z. Wang, Phys. Rev. Lett. 121, 086803 (2018). tinuum bands, and applications of the non-Bloch band the-
84
S. Yao, F. Song, and Z. Wang, ory to some models, which includes Refs. 74, 78, 83, 95, 97–
Phys. Rev. Lett. 121, 136802 (2018). 99.
85 97
K. Kawabata, K. Shiozaki, and M. Ueda, S. Ryu and Y. Hatsugai,
Phys. Rev. B 98, 165148 (2018). Phys. Rev. Lett. 89, 077002 (2002).
86 98
C. Yuce and Z. Oztas, Sci. Rep. 8, 17416 (2018). D. C. Brody, J. Phys. A 47, 035305 (2014).
87 99
K. Kawabata, S. Higashikawa, Z. Gong, Y. Ashida, and S. Ryu, A. P. Schnyder, A. Furusaki, and A. W. Ludwig,
M. Ueda, Nat. Commun. 10, 297 (2019). New J. Phys, 12, 065010 (2010).
 Supplemental Material for “Non-Bloch Band Theory of Non-Hermitian Systems”
Kazuki Yokomizo1 and Shuichi Murakami1, 2
1
Department of Physics, Tokyo Institute of Technology,
2-12-1 Ookayama, Meguro-ku, Tokyo, 152-8551, Japan
2
TIES, Tokyo Institute of Technology, 2-12-1 Ookayama, Meguro-ku, Tokyo, 152-8551, Japan
arXiv:1902.10958v2 [cond-mat.mes-hall] 12 Aug 2019

SI. SIMPLE NON-HERMITIAN +N +N

TIGHT-BINDING MODEL

1 q 1 q 1 q
In this section, we study a simple non-Hermitian tight-
binding model in terms of the non-Bloch band theory. FIG. S1. Schematic figure of a 1D tight-binding system. A
The Hamiltonian of this system is given by unit cell includes q degrees of freedom, and the range of hop-
ping is N .
X
L−1 
H= tR c†n+1 cn + tL c†n cn+1 , (S1)
n=1
Then we can get the discrete energy levels in a finite open
where tR , tL ∈ R are the asymmetric hopping amplitudes, chain.
and L is the system size. We put tR and tL to be positive, As the system size L becomes larger, these energy lev-
T
for simplicity. For the eigenvector |ψi = (ψ1 , · · · , ψL ) , els becomes dense, and finally, the energy spectrum can
the real-space eigen-equation H |ψi = E |ψi can be writ- be obtained in the thermodynamic limit L → ∞. These
ten as continuum bands at L → ∞ can be obtained by setting
Eq. (S6). Namely Eq. (S6) leads to
tR ψn−1 + tL ψn+1 = Eψn , (n = 1, · · · , L). (S2)
r r
tR iθ tR −iθ
Henceforth we imply ψ0 = ψL+1 = 0 as a boundary β1 = e , β2 = e , (S9)
condition. From the theory of linear difference equations, tL tL
a general solution of the recursion equation Eq. (S2) is
and from Eq. (S4), we get
written as
√
n n E = 2 tR tL cos θ, (S10)
ψn = (β1 ) φ(1) + (β2 ) φ(2) , (S3)

where βj (j = 1, 2) are the solutions of the eigenvalue giving the continuum bands by changing θ ∈ R. There-
equation fore we can interpret Eq. (S6) as the condition for the
continuum bands.
tR βj−1 + tL βj = E. (S4) In Eq. (S8), apart from the factor rn , the wave function
(m)
ψn represents a standing wave, as a superposition of
Here the boundary condition ψ0 = 0 gives φ(1) + φ(2) =
two counterpropagating plane waves. Thus the condition
0. By combining this equation, the boundary condition
L+1 for the continuum bands Eq. (S6) can be also understood
ψL+1 = 0 gives (β1 /β2 ) = 1, and we can rewrite this as a condition for formation of a standing wave in this
equation as system.
 
β1 2iθm mπ
=e , θm = , m = 1, · · · , L . (S5)
β2 L+1
SII. NON-BLOCH BAND THEORY
Then we get
In this section, we describe a systematic way to get
|β1 | = |β2 | , (S6) the generalized Brillouin zone Cβ in general cases. In
p particular, Cβ becomes a unit circle in Hermitian cases.
and we obtain r ≡ |β1,2 | = tR /tL since β1 β2 = tR /tL
from Eq. (S4). Therefore βj (j = 1, 2) can be given by
(m) (m) A. General recipe for deriving the generalized
β1 = reiθm , β2 = re−iθm , (S7)
Brillouin zone
and the eigenstate Eq. (S3) and the eigenenergy Eq. (S4)
can be written as We start with a one-dimensional (1D) tight-binding

n
n model in an open chain as shown in Fig. S1. A unit cell
ψn(m) ∝ reiθm − re−iθm ∝ rn sin nθm ,
√ is composed of q degrees of freedom, such as sublattices,
E (m) = 2 tR tL cos θm . (S8) spins, or orbitals, and the range of hopping is N , meaning
 2

that electrons hop up to the N -th nearest unit cells. Then SSH model, Cβ is shown in Fig. S4 (a). We note that
its Hamiltonian can be written as Eq. (S17) was proposed in Ref. 1 in a simpler case of
q
Eq. (S15) being a quadratic equation for β correspond-
N
X X X ing to M = 1, but for general cases its condition has
H= ti,µν c†n+i,µ cn,ν , (S11)
not been known so far. We also mention Ref. 2, which
n i=−N µ,ν=1
discusses the condition for continuum bands for limited
cases of M = 1, JL = JR = J, and J 2 = 0, where JL
where c†n,µ (cn,µ ) is a creation (an annihilation) oper-
and JR are matrices formed by t1,µν and t−1,µν in our
ator of an electron with index µ (µ = 1, · · · , q) in the
notation.
n-th unit cell, and ti,µν is a hopping amplitude to the
i-th nearest unit cell. This Hamiltonian can be non- We here mention special cases in which the degree of
Hermitian, meaning that ti,µν is not necessarily equal to the eigenvalue equation Eq. (S15) is less than 2M = 2qN .
t∗−i,νµ . Here one can write the real-space eigen-equation This occurs when some hopping amplitudes are zero.
as H |ψi = E |ψi, where the eigenvector is written as Even in these cases, one can restore the eigenvalue equa-
T tion to an algebraic equation with the degree 2M for β,
|ψi = (· · · , ψ1,1 , · · · , ψ1,q , ψ2,1 , · · · , ψ2,q , · · · ) . Thanks
by including all hopping amplitudes to the N -th nearest
to the spatial periodicity, we can write the eigenvector as
unit cells to be non-zero, through an addition of infinites-
a linear combination:
imally small hopping amplitudes to the otherwise zero
X n (j) hopping amplitudes. Then Eq. (S17) can be applied in
ψn,µ = φ(j) (j)
n,µ , φn,µ = (βj ) φµ , (µ = 1, · · · , q).
this situation. By taking the zero limit of all these in-
j
(S12) finitesimal hopping amplitudes, continuum bands in the
(j) original system can be obtained. In Sec. SIV B, we de-
Then, by imposing that φn,µ is an eigenstate, the eigen- scribe such special cases, and we show how to apply our
equation can be rewritten for β = βj as theory to them.
q
X
[H (β)]µν φν = Eφµ , (µ = 1, · · · , q) , (S13)
ν=1 B. Hermitian cases
where a generalized Bloch Hamiltonian H (β) is given by
In this subsection, we prove that the generalized Bril-
N
X louin zone Cβ becomes a unit circle in Hermitian cases.
[H (β)]µν = ti,µν β i , (µ, ν = 1, · · · , q) . (S14) In the following, we assume that the eigenvalue equation
i=−N Eq. (S15) is an algebraic equation for β of 2M -th degree.
In this case, it can be written as
We note that in Hermitian cases, Eq. (S13) is nothing but
the eigen-equation for the Bloch eigenstates. Therefore M
X
we can get the eigenvalue equation as ai β i = 0, (S18)
i=−M
det [H (β) − E] = 0. (S15)
where ai (i = −M, · · · , M ) are coefficients satisfying
We note that the eigenvalue equation is an algebraic
a−i = a∗i and a0 is real. These coefficients are func-
equation for β with an even degree 2M = 2qN in general
tions of the eigenenergy E and the hopping terms ti,µν
cases, where M = qN because each element of the q × q
included in Eq. (S11). Here the eigenenergy E is real
matrix H (β) has the form of Eq. (S14).
due to Hermiticity of the Hamiltonian. Equation (S18)
Next we show how to determine the generalized Bril-
has 2M solutions, and they are numbered so as to satisfy
louin zone Cβ , which determines continuum bands. It
Eq. (S16). Since one can rewrite Eq. (S18) as
is one of the main results of our work. Let βj (j =
1, · · · , 2M ) be the solutions of the eigenvalue equation M M
Eq. (S15). Here we number the 2M solutions so as to X ∗ −i
X −i
(a−i ) (β ∗ ) = ai (β ∗ ) = 0, (S19)
satisfy
i=−M i=−M

|β1 | ≤ |β2 | ≤ · · · ≤ |β2M−1 | ≤ |β2M | . (S16)

the solutions always appear in pairs: (β, 1/β ∗ ), and we
We find that the condition to get continuum bands can get β2M+1−j = 1/βj∗ . In particular, one can get the
∗
be written as relationship between βM and βM+1 as βM+1 = 1/βM .
Therefore Eq. (S17) can be rewritten as
|βM | = |βM+1 | , (S17)
|βM | = |βM+1 | = 1. (S20)
and then the trajectory of βM and βM+1 satisfying
Eq. (S17) determines Cβ . The derivation of Eq. (S17) Thus we can conclude that Cβ becomes a unit circle in
is given in Sec. SIII. For example, in the non-Hermitian Hermitian systems.
 3

SIII. DERIVATION OF THE CONDITION FOR see how the solutions for a large system size form contin-
CONTINUUM BANDS uum bands. Therefore we suppose L to be quite large,
and consider a condition to form densely distributed en-
In this section, we derive the condition for continuum ergy levels. Here Eq. (S24) is expressed in a form
bands Eq. (S17). To this end, we focus on boundary X Y L
conditions in a finite open chain with L unit cells. The F (βi∈P , βj∈Q , E, S) (βk ) = 0. (S25)
wave functions are written in the form Eq. (S12), i.e. P,Q k∈P

2M
X The sets P and Q are two disjoint subsets of the set
n
ψn,µ = (βj ) φ(j)
µ , (n = 1, · · · L, µ = 1, · · · , q). {1, · · · , 2M } such that the number of elements of each
j=1 subset is M . In this way, the sum included in Eq. (S25)
(S21) is taken over all the sets P and Q.
The equations from a boundary condition include the We now consider asymptotic behavior of the solutions
(j)
2qM unknown variables φµ . The real-space eigen- of Eq. (S25) for large L. When |βM | 6= |βM+1 |, there is
L
equation H |ψi = E |ψi fixes the ratio between the values only one leading term proportional to (βM+1 · · · β2M )
(j)
of φµ sharing the same value of j. Therefore one can in Eq. (S25) in the limit of a large L. Thus it leads to
(j)
reduce the 2qM variables to the 2M variables φµ with the equation
a single value of µ, e.g. µ = 1. As a result, we can get
(j) F (βi∈P ′ , βj∈Q′ , E, S) = 0 (S26)
M equations for the 2M variables φ1 (j = 1, · · · , 2M )
at the left end of the open chain around n = 1:
with P ′ = {M + 1, · · · , 2M } and Q′ = {1, · · · , M }. This
2M equation does not depend on L, and does not allow
X (j)
fi (βj , E, S) φ1 = 0, (i = 1, · · · , M ) , (S22) continuum bands. On the other hand, when |βM | =
j=1 |βM+1 |, there are two leading terms proportional to
L L
(βM βM+2 · · · β2M ) and to (βM+1 βM+2 · · · β2M ) , and
and M equations at the right end of the open chain the equation is written as
around n = L:
 L
βM F (βi∈P0 , βj∈Q0 , E, S)
2M
X =− (S27)
gi (βj , E, S) (βj ) L (j)
φ1 = 0, (i = 1, · · · , M ) , (S23) βM+1 F (βi∈P1 , βj∈Q1 , E, S)
j=1
with P0 = {M + 1, · · · , 2M }, Q0 = {1, · · · , M }, P1 =
where S is the set of the system parameters ti,µν . Here {M, M + 2, · · · , 2M }, and Q1 = {1, · · · , M − 1, M + 1}.
fi and gi are functions of βj , E, and S, and they do In such a case, we can expect that the relative phase be-
not depend on L. We emphasize here that any kinds of tween βM and βM+1 can be changed almost continuously
boundary conditions for an open chain can be written as for a large L, producing the continuum bands.
M equations (Eq. (S22)) for the left end and M equations In Sec. SIV, we derive Eq. (S17) for the non-Hermitian
(Eq. (S23)) for the right end. By combining Eqs. (S22) SSH model as an example. In this model, Eq. (S25) com-
and (S23), we can obtain a condition for the existence of ing from the boundary condition is given by Eq. (S40).
(1) (2M)
nontrivial solutions for φ1 , · · · , φ1 :

f1 (β1 , E, S) ··· f1 (β2M , E, S) SIV. NON-HERMITIAN SSH MODEL

.. .. ..
. . .
In this section, we show how to get the eigenvalue equa-
fM (β1 , E, S) ··· fM (β2M , E, S) tion and the condition for the continuum bands in the
L L = 0. non-Hermitian SSH model. We show that even when an
g1 (β1 , E, S) (β1 ) · · · g1 (β2M , E, S) (β2M )
.. .. .. arbitrary complex potential is added on the boundary
. . . sites, the condition for the continuum bands Eq. (S17)
L L is unchanged, and namely the generalized Brillouin zone
gM (β1 , E, S) (β1 ) · · · gM (β2M , E, S) (β2M )
is unchanged. Thus we emphasize that the condition
(S24) for continuum bands, Eq. (S17), is independent of any
boundary conditions.
We note that both fi and gi depend on boundary condi-
tions but not on L.
The determinant included in Eq. (S24) is an algebraic
equation for βj (j = 1, · · · , 2M ). By solving Eq. (S24), A. Eigenvalue equation
one can calculate eigenenergies of the system with open
boundaries. One cannot analytically solve these equa- In this subsection, we derive the eigenvalue equation in
tions for a general system size. Nonetheless our aim is to the non-Hermitian SSH model as shown in Fig. S2. This
 4

FIG. S2. Schematic figure of the non-Hermitian SSH model.

The dotted boxes indicate the unit cell. FIG. S3. Schematic figure of the non-Hermitian SSH model
with the complex potential a and b on the boundary sites. We
show that the number of unit cells is three in this figure.
model is the case of q = 2 and N = 1, giving M = 2.
The Hamiltonian can be written as
X h Henceforth we assume t2 6= ±γ2 /2 and t3 6= 0, meaning
γ1  †  γ1  †
H= t1 + cn,A cn,B + t1 − cn,B cn,A that Eq. (S34) is a quartic equation for β, having four
n
2 2 solutions βj (j = 1, · · · , 4). They are numbered so as to
 γ2  †  γ2  † satisfy |β1 | ≤ |β2 | ≤ |β3 | ≤ |β4 |.
+ t2 + cn,B cn+1,A + t2 − cn+1,A cn,B
 2 i 2
+t3 c†n,A cn+1,B + c†n+1,B cn,A , (S28) B. Derivation of the boundary equation and the
condition for the continuum bands
where t1 , t2 , t3 , γ1 , and γ2 are real. Eigenvectors are writ-
T
ten as |ψi = (· · · , ψ1,A , ψ1,B , ψ2,A , ψ2,B , · · · ) . Then we
Next we study the non-Hermitian SSH model with L
can explicitly write the equation for ψn,µ as
 unit cells with the complex potential a and b on the
γ2   γ1  boundary sites as shown in Fig. S3. In this subsection,
t2 − ψn−1,B + t1 + ψn,B + t3 ψn+1,B = Eψn,A ,
2   2  we show that the condition for the continuum bands is
γ1 γ2  indeed given by Eq. (S17), independent of a, b, and L.
t3 ψn−1,A + t1 − ψn,A + t2 + ψn+1,A = Eψn,B .
2 2 We write down the boundary condition for ψn,µ :
(S29)  γ1 
t1 + ψ1,B + t3 ψ2,B + aψ1,A = Eψ1,A ,
Here we take an ansatz for the wave function as a linear  2 
combination: γ1 γ2 
t1 − ψ1,A + t2 + ψ2,A = Eψ1,B ,
X  2  2
ψn,µ = φ(j)
n,µ , (µ = A, B), (S30) γ2 γ1
t2 − ψL−1,B + t1 + ψL,B = EψL,A ,
j 2   2
γ1
(j) (j) t3 ψL−1,A + t1 − ψL,A + bψL,B = EψL,B .(S35)
where φn,A and φn,B takes the exponential form 2
    Then, by substituting the general solution
(j) (j) n (j) (j)
φn,A , φn,B = (βj ) φA , φB . (S31) 4
X n
ψn,µ = (βj ) φ(j)
µ , (µ = A, B) (S36)
By substituting the exponential form Eq. (S31) into j=1
Eq. (S29), one can obtain the bulk eigen-equations:
h to Eq. (S35), one can obtain four equations for the eight
γ2  −1  γ1  i
(j)
t2 − β + t1 + + t3 β φB = EφA , coefficients φµ (j = 1, · · · , 4, µ = A, B). By recalling
h 2 2 that these coefficients satisfy
γ1   γ2  i
t3 β −1 + t1 − + t2 + β φA = EφB , E
2 2 (j)
φA =
(j)
φ ,
(S32) t3 βj + (t1 − γ1 /2) + (t2 + γ2 /2) βj B
−1

and the generalized Bloch Hamiltonian H (β): (j) E (j)

φB = φ ,
(t2 − γ2 /2) βj−1 + (t1 + γ1 /2) + t3 βj A
H (β) = R+ (β) σ+ + R− (β) σ− ,
(S37)
R+ (β) = (t2 − γ2 /2) β −1 + (t1 + γ1 /2) + t3 β,
from the bulk eigen-equation Eq. (S32), we can reduce
R− (β) = t3 β −1 + (t1 − γ1 /2) + (t2 + γ2 /2) β −1 ,
the problem into four linear equations for the four coef-
(S33) (j)
ficients φA (j = 1, · · · , 4). Then the condition for the
where σ± = (σx ± iσy ) /2. Therefore the eigenvalue linear equation to have nontrivial solutions is written as
equation can be written as 1 1 1 1
h γ2  −1  γ1  i
t2 − β + t1 + + t3 β f1 f2 f3 f4
2  2 = 0, (S38)
h γ1   γ2  i X1 β1L X2 β2L X3 β3L X4 β4L
−1
× t3 β + t1 − + t2 + β = E 2 . (S34) g1 β1L g2 β2L g3 β3L g4 β4L
2 2
 5

where Xj , fj and gj (j = 1, · · · , 4) are defined as (a) | b2| = | b3| (b) |bi| = | b j|

Im(b ) Im(b)
βj
Xj = −1 ,
(t2 − γ2 /2) βj + (t1 + γ1 /2) + t3 βj
fj = aβj − (t2 − γ2 /2) Eβj−1 Xj ,
gj = bEβj−1 Xj − (t2 + γ2 /2) βj , (S39) Re(b) Re(b)

respectively. Furthermore Eq. (S38) can be explicitly

written as
1X

L
sgn (σ) fσ(1) − fσ(2) gσ(3) Xσ(4) βσ(3) βσ(4) = 0, FIG. S4. Trajectories of β. (a) Trajectory of β2 and β3
2 σ
with |β2 | = |β3 | and (b) trajectories of βi and βj with
(S40) |βi | = |βj | (i 6= j). (a) corresponds to the generalized Bril-
louin zone Cβ . The parameters are same as Fig. 3 (c-1) in
where the sum is taken over all the permutations σ for the main text: t1 = 0.3, t2 = 0.5, t3 = 1/5, γ1 = 5/3, and
four objects. γ2 = 1/3.
Next we demonstrate that the condition for the contin-
uum bands is given by |β2 | = |β3 |. We have to consider
the condition when the solutions of Eq. (S40) are densely
distributed for a large L. When |β2 | 6= |β3 |, the only lead- Another special case is the case of t3 = 0 in the non-
L
ing term in Eq. (S40) is the term proportional to (β3 β4 ) , Hermitian SSH model. In this case, the eigenvalue equa-
leading to an equation (f1 − f2 ) (g3 X4 − g4 X3 ) = 0 in tion becomes a quadratic equation for β, and it has two
the thermodynamic limit L → ∞. By combining this solutions. Even in this special case, if the infinitesimally
equation with the eigenvalue equation (S34), the eigenen- small hopping amplitude t3 is added, the eigenvalue equa-
ergies are restricted to discrete values, and it cannot rep- tion has the four solutions β = β1 , β2 , β3 , β4 satisfying
resent the continuum bands. On the other hand, when Eq. (S16). Then the above condition corresponds to the
condition β1 (= 0) ≤ |β2 | = |β3 | ≤ β4 (= ∞) in the limit
|β2 | = |β3 | (S41)
t3 → 0, meaning that the two solutions (β2 , β3 ) of the
L
is satisfied, Eq. (S40) has two leading terms, (β2 β4 ) and quadratic equation have the same absolute values. This
L result agrees with the previous work1 .
(β3 β4 ) for a large L, and Eq. (S40) can be rewritten as
 L
β2 (f1 − f2 ) (g4 X3 − g3 X4 )
= . (S42)
β3 (f1 − f3 ) (g4 X2 − g2 X4 )
For a large L, this equation allows a dense set of solutions
when the relative phase between β2 and β3 is continu- C. Generalized Brillouin zone
ously changed. Therefore we conclude that Eq. (S41) is
an appropriate condition for the continuum bands. We As an example, we show the generalized Brillouin zone
here emphasize that Eq. (S41) is now independent of any Cβ for parameters t1 = 0.3, t2 = 0.5, t3 = 1/5, γ1 = 5/3,
boundary conditions a and b, and the system size L. If we and γ2 = 1/3 in our model. We impose the condition
change the value of the parameters, Eq. (S40) changes; |β2 | = |β3 | and calculate Cβ as the trajectories of β2 and
nonetheless Eq. (S41) remains the same, and together β3 shown in Fig. S4 (a). For comparison, we investi-
with the eigenvalue equation, we can obtain the contin- gate what happens if we impose a condition |βi | = |βj |
uum bands. for some i and j among the four solutions instead. We
With a special choice of parameters, the degree of show the trajectory of βi and βj in Fig. S4 (b) under
the eigenvalue equation may become less than 2M . For the condition |βi | = |βj | for some i and j. In Ref. 1, it
example, when t2 = −γ2 /2, Eq. (S34) becomes a cu- was suggested that |βi | = |βj | is a necessary condition
bic equation for β. In this case, we can still regard for the continuum bands. Nonetheless it is not sufficient
Eq. (S34) as a quartic equation by putting t2 + γ2 /2 to to get Cβ , and we should restrict this condition to be
be an infinitesimal. Therefore, by adding another solu- |β2 | = |β3 |, leading to Cβ in Fig. S4 (a).
tion β = ∞ to three solutions for the cubic equation, our
result Eq. (S41) holds good with four solutions β1 , β2 , In some cases, Cβ can have cusps as seen in Fig. S4 (a).
β3 , β4 (= ∞). Thus, in general, when the degree of the It appears when three of the four solutions of β share the
eigenvalue equation is less than 2M , one can treat it as same absolute value. Suppose |β1 | < |β2 | = |β3 | < |β4 |,
a limiting case of an algebraic equation of 2M -th degree and as we go along Cβ , |β1 | approaches |β2 | (= |β3 |).
by formally adding a solution β = 0 or β = ∞, depend- Then, when |β1 | = |β2 | = |β3 |, the behavior of the so-
ing on the system, and the condition for the continuum lutions satisfying |β2 | = |β3 | changes, and there appears
bands remains valid. a cusp in Cβ , as one can see from Figs. S4 (a) and (b).
 6

D. Methods for computing the generalized where 1 is an N ×N identity matrix. Then the eigenvalue
Brillouin zone equations for the right and left eigenstates
!
We explain our method to get the trajectory of β sat- |aR (β)i
isfying the continuum-band condition (S41), as shown |ψR (β)i = ,
in Fig. S4. We first express the eigenvalue equation |bR (β)i


Eq. (S34) as E 2 = f (β). Suppose the two solutions β hψL (β)| = haL (β)| , haR (β)| , (S48)
and β ′ have the same absolute values: |β| = |β ′ |. Then
we have are given by
β ′ = βeiθ , (S43)
where θ is real. Then, by taking the difference between R+ (β) |bR (β)i = E |aR (β)i ,
two equations: R− (β) |aR (β)i = E |bR (β)i , (S49)


E 2 = f (β) , E 2 = f βeiθ , (S44)
and
we get
 γ2 

0 = t2 − t3 β −2 1 − e−2iθ haL (β)| R+ (β) = E hbL (β)| ,
h 2
γ1   γ2   γ1  i −1
hbL (β)| R− (β) = E haL (β)| , (S50)
+ t1 − t2 − + t1 + t3 β 1 − e−iθ
h 2 2 2
γ1   γ1   γ2 i
respectively. For the right and left eigenstates, one can
+ t1 − t3 + t1 + t2 + β 1 − eiθ
 2 2 2 reduce Eqs. (S49) and (S50) to
γ2  2 2iθ


+ t2 + t3 β 1 − e . (S45)
2 R+ (β) R− (β) |aR (β)i = E 2 |aR (β)i ,
This equation allows us to compute β for a given value
of θ ∈ (0, 2π). Then we obtain a set of values of β that haL (β)| R+ (β) R− (β) = E 2 haL (β)| , (S51)
satisfies |β| = |β ′ |. Here, to obtain the generalized Bril-
louin zone Cβ , we should further constrain the values of respectively. Here we introduce the right and
β by Eq. (S41). Namely the absolute values of β and left eigenstates of the N × N matrix R+ (β) R− (β)
β ′ should be the second and third largest ones among as |aR/L,1 (β)i , · · · , |aR/L,N (β)i, respectively, and the
the four solutions. By selecting the values of β and β ′ eigenvalues as E12 (β) , · · · , EN2
(β). Furthermore the
satisfying this condition, we can get Cβ . right and left eigenstates satisfy

haL,i (β) | aR,j (β)i = δij (S52)

SV. Q MATRIX AND WINDING NUMBER IN
1D NON-HERMITIAN SYSTEMS WITH CHIRAL
SYMMETRY since one can take the biorthogonal basis3,4 . Therefore
we can obtain the biorthogonal eigenstates of the Hamil-
A. Multi-band model tonian Eq. (S46) in the occupied bands for i = 1, · · · , N
as
In this section, we focus on 1D non-Hermitian systems  
with chiral symmetry which have an arbitrary number |aR,i (β)i
1  ,
of bands, and derive a formula for the winding number. |ψR,i (β)i = √ 1
2 − R− (β) |aR,i (β)i
In the following, the systems have a gap around E = 0, Ei (β)
but without exceptional points. Since the systems always  
1 1
have pairs of the eigenenergy (E, −E) due to chiral sym- hψL,i (β)| = √ haL,i (β)| , − haL,i | R+ (β) .
2 Ei (β)
metry, we can assume that the bands are composed of
N occupied bands with E = −Ei (i = 1, · · · , N ) and N (S53)
unoccupied bands with E = Ei (i = 1, · · · , N ). By tak-
ing an appropriate basis, the Hamiltonian in the systems Then the unoccupied eigenstates are given by
can be written as the block off-diagonal form: |ψeR/L,i (β)i ≡ σz |ψR/L,i (β)i. In the systems, the
!
0 R+ (β) Q matrix Q (β) can be defined as
H (β) = , (S46)
R− (β) 0 N 
X 
where β ≡ eik , k ∈ C and R± (β) are N × N matrices. Q (β) = |ψeR,i (β)i hψeL,i (β)| − |ψR,i (β)i hψL,i (β)|
The chiral symmetry is expressed as
i=1
!
1 0 (S54)
σz H(β) = −H(β)σz , σz ≡ , (S47) and in the matrix form, it can be written as
0 −1
 7

N
!
X 1 O |aR,i (β)i haL,i (β)| R+ (β)
Q (β) = . (S55)
i=1
Ei (β) R− (β) |aR,i (β)i haL,i (β)| O

p
Here we define the matrix q (β) and q −1 (β) as Therefore we obtain q = R+ / R+ R− , and the winding
number w as
Z
N
X 1 i
q (β) ≡ |aR,i (β)i haL,i (β)| R+ (β) , w= dq q −1 (β)
Ei (β) 2π Cβ
i=1 Z
N
i
X 1 = d log q (β)
q −1 (β) = R− (β) |aR,i (β)i haL,i (β)| , 2π Cβ
i=1
Ei (β) 1
=− [arg q (β)]Cβ
(S56) 2π
1 [arg R+ (β) − arg R− (β)]Cβ
respectively. As a result, one can get the winding number =− . (S62)
2π 2
w1,5 as Thus the winding number w is determined by the
Z change of the phase of R± (β) when β goes along the
i  
w= Tr dq q −1 (β) . (S57) generalized Brillouin zone Cβ . On a complex plane, let
2π Cβ ℓ± denote the loops drawn by R± (β) when β goes along
Cβ in the counterclockwise way. Then w is determined
by the number of times that ℓ± surround the origin O.
When neither ℓ+ nor ℓ− surrounds O, w is zero. It takes
B. Two-band model
a non-zero value when they simultaneously surround O.
Here we can show the bulk-edge correspondence in this
In particular, we can explicitly write Eq. (S57) in a case. If parameters of the system can be continuously
two-band model. The Hamiltonian can be written as changed without closing the gap, the winding number
w does not change, and the topology of the systems re-
H (β) = R+ (β) σ+ + R− (β) σ− , (S58) mains invariant. Namely, if the systems after changing
the values of the parameters have the zero-energy edge
where σ± = (σx ± iσy ) /2. Then the eigenvalues are states, one can conclude that the original systems also
given by have the zero-energy edge states. This can be proved
even in non-Hermitian cases, following the proof in Her-
p mitian systems6 .
E± (β) = ± R+ (β) R− (β). (S59) Suppose we change the parameters of the system con-
tinuously, and at some values of the parameters, ℓ+ and
The right and left eigenstates can be written down as ℓ− simultaneously pass the origin O on the R plane.
! At this time, the gap closing occurs because the energy
1 R+ eigenvalues are given by Eq. (S59).
|uR,+ i = √ p p , In Hermitian systems, R+ ∗
= R− holds, and two loops
2 R+ R− R+ R−
! ℓ+ and ℓ− are related by complex conjugation, so
1 R+
|uR,− i = √ p p , [arg R− (β)]Cβ = − [arg R+ (β)]Cβ . (S63)
2 R+ R− − R+ R−
1 p
In contrast with these cases, in some non-Hermitian sys-
huL,+ | = √ p R− , R+ R− , tems, only one of the two loops ℓ+ and ℓ− passes the
2 R+ R− origin O. In that case, only one of the two values R+
1 p
or R− become 0, leading to E = 0, and the Hamiltonian
huL,− | = √ p R− , − R+ R− , (S60)
2 R+ R− is written as the Jordan normal form. It represents that
the system has exceptional points. In Figs. 3 (a) and
respectively. The subscript +/− means that the eigen- 4 (a) in the main text, the phase with the exceptional
states with +/− have the eigenvalues E+ or E− , respec- point extends over a finite region due to the following
tively. From Eq. (S60), the Q matrix Q (β) is given by reason. When R+ = 0 (or R− = 0) has non-real solu-
tions, these two solutions β1+ and β2+ (or β1− and β2− )
should be complex conjugate, having the same absolute
Q (β) = |uR,+ (β)i huL,+ (β)| − |uR,− (β)i huL,− (β)|
  value, because all the coefficients of R+ = 0 and R− = 0
1 0 R+ (β) are real. This persists as long as the following condi-
= p . (S61)
R+ (β) R− (β) R− (β) 0 tion are satisfied: R+ = 0 (or R− = 0) has non-real
 8

(j) (j)
where φn,A and φn,B take the exponential form
   
(j) (j) (j) (j)
φn,A , φn,B = (βj )n φA , φB . (S67)

By substituting the exponential form Eq. (S67) into

Eq. (S65), one can obtain
  hr i
ir −1

iγ

− β−β + φA + β + β −1 + v φB = EφA ,
2 2 2
FIG. S5. (a) Schematic figure of the tight-binding model pro-   hr i
ir
iγ

posed in the previous work7 . This model includes the gain on β − β −1 − φB + β + β −1 + v φA = EφB ,
sublattice A and the loss on sublattice B. (b) Energy bands 2 2 2
in a finite open chain of the model given by Eq. (S64). We (S68)
set the parameter as r/γ = 1/3 and the number of √ unit cells
as L = 50. The√gap closes at v/γ = ±c1 = ± 13/6 and and the generalized Bloch Hamiltonian H (β):
v/γ = ±c2 = ± 5/6. The red dashed lines represent the  
h r
i ir
iγ
gap-closing points in the continuum bands.
H (β) = v + β + β −1 σx + − β − β −1 + σz .
2 2 2
two solutions satisfying β1− < β1+ = β2+ < β2− (or (S69)
β1+ < β1− = β2− < β2+ ). We note that the wind- Therefore the eigenvalue equation is given by
ing number is not well-defined when the system has the   
exceptional pints because the gap is closed. γ γ2  γ  −1
r v+ β + r2 + v 2 − − E2 + r v − β = 0.
2 4 2
(S70)
SVI. CALCULATION ON ANOTHER MODEL
The condition for the continuum bands can be written
In this section, we investigate the 1D non-Hermitian as |β1 | = |β2 | since Eq. (S70) is a quadratic equation for
tight-binding model proposed in Ref. 7 as shown in β. In this case, the generalized Brillouin zone becomes a
Fig. S5 (a) in terms of the complex Bloch wave number circle with the radius
k ∈ C. s
First of all, we can write the Hamiltonian in this model v − γ/2
R ≡ |β1,2 | = (S71)
as v + γ/2
Xh  † 
H= v cn,A cn,B + c†n,B cn,A which is given by Vieta’s formula1 . By substituting β =
n iϕ
ir  † Re , ϕ ∈ R to Eq. (S70), we can get the gap-closing
+ c † † †
cn,A − cn,A cn+1,A − cn+1,B cn,B + cn,B cn+1,B point as
2  n+1,A  s
r †  2
+ cn+1,A cn,B + c†n,B cn+1,A + c†n+1,B cn,A + c†n,A cn+1,B v 1 r
2 = ± ± . (S72)
   γ 4 γ
iγ †
+ cn,A cn,A − c†n,B cn,B , (S64)
2 Therefore the zero-energy edge states are ex-
where we take r, v, and γ to be real, and hence- pected to  appear q in the q parameter  region
forth, we set |r/γ| ≤ 1/2. When the eigenvector v/γ ∈
2
− 1/4 + (r/γ) , − 1/4 − (r/γ)
2
and
of the real-space eigen-equation is written as |ψi = q q 
T
(· · · , ψ1,A , ψ1,B , ψ2,A , ψ2,B , · · · ) , the equation for ψn,µ v/γ ∈
2
1/4 − (r/γ) , 1/4 + (r/γ)
2
because
can be written as
ir iγ the winding number is shown to be unity there, as
(ψn−1,A − ψn+1,A ) + ψn,A a consequence of the previous discussion. Indeed we
2 2
r confirm the appearance of these zero-energy edge states
+ (ψn−1,B + ψn+1,B ) + vψn,B = Eψn,A , in a finite open chain with r/γ√= 1/3 within these region.
2 √
ir iγ The gap closes at v/γ = ± 5/6, ± 13/6 in the con-
− (ψn−1,B − ψn+1,B ) − ψn,B tinuum bands,
 √ and the
√ zero-energy edge states
√ √appear
2 2
r in v/γ ∈ − 13/6, − 5/6 and v/γ ∈ 5/6, 13/6 ,
+ (ψn−1,A + ψn+1,A ) + vψn,A = Eψn,B . (S65) which perfectly agrees with our conclusion of energy
2
levels in a finite open chain (Fig. S5 (b)). Thus, by our
Here we take an ansatz as a linear combination:
X non-Bloch band theory, the bulk-edge correspondence
ψn,µ = φ(j)
n,µ , (µ = A, B), (S66) is fully shown in the same sense as in the Hermitian
j systems. Namely the closing of the gap in the open
 9

chain is explained by Eq. (S72). At these points, the tary transformation

winding number defined in our theory changes between
zero and unity, which exactly coincides with absence σx → σx , σy → −σz , σz → σy , (S73)
and presence of zero-energy in-gap states. These aspects
have not been understood within theories with the real the generalized Bloch Hamiltonian Eq. (S69) can be
Bloch wave number7 . rewritten as
 γ   γ 
H (β) → v + + rβ −1 σ+ + v − + rβ σ− , (S74)
2 2
where σ± = (σx ± iσy ) /2. We note that the asymmetric
hopping amplitude is caused by the gain and loss in the
original system. This Hamiltonian is reduced to the non-
The tight-binding model in Fig. S5 (a) can be regarded Hermitian SSH model as shown in Fig. S2 with t3 = γ2 =
as the non-Hermitian SSH model1 . Indeed, by the uni- 0.

1 5
S. Yao and Z. Wang, Phys. Rev. Lett. 121, 086803 (2018). S. Ryu, A. P. Schnyder, A. Furusaki, and A. W. Ludwig,
2
F. K. Kunst and V. Dwivedi, New J. Phys, 12, 065010 (2010).
6
Phys. Rev. B 99, 245116 (2019). S. Ryu and Y. Hatsugai,
3
D. C. Brody, J. Phys. A 47, 035305 (2014). Phys. Rev. Lett. 89, 077002 (2002).
4 7
H. Shen, B. Zhen, and L. Fu, T. E. Lee, Phys. Rev. Lett. 116, 133903 (2016).
Phys. Rev. Lett. 120, 146402 (2018).