Commun. Math. Phys.

203, 341 – 347 (1999) Communications in

Mathematical
Physics
© Springer-Verlag 1999

On the Absolutely Continuous Spectrum

of One-Dimensional Schrödinger Operators
with Square Summable Potentials
P. Deift1 , R. Killip2
1 Courant Institute, 251 Mercer St., New York, NY 10012, USA. E-mail: deift@cims.nyu.edu
2 Mathematics 253-37, Caltech, Pasadena, CA 91125, USA. E-mail: killip@cco.caltech.edu

Received: 30 October 1998 / Accepted: 16 November 1998

Abstract: For continuous and discrete one-dimensional Schrödinger operators with

square summable potentials, the absolutely continuous part of the spectrum is essentially
supported by [0, ∞) and [−2, 2] respectively. This fact is proved by considering a priori
estimates for the transmission coefficient.

1. Introduction
In this paper we consider the absolutely continuous spectrum of one-dimensional Schrö-
dinger operators with potentials decaying slowly at infinity. For example, we consider
the half-line operators with Dirichlet boundary conditions:
d 2ψ
HD ψ(x) = − + q(x)ψ(x) on L2 ([0, ∞)) with ψ(0) = 0 (1)
dx 2
and

ψ(n + 1) + ψ(n − 1) + q(n)ψ(n) n = 1, 2, . . .
hD ψ(n) = . (2)
ψ(1) + q(0)ψ(0) n=0
Recent papers by Christ, Kiselev and Remling [1,2,11] have shown that for potentials
obeying |q(x)| ≤ C(1 + |x|)−1/2− ,  > 0, the a.c. spectrum of HD (resp. hD ) is essen-
tially supported by [0, ∞) (resp. [−2, 2]). By essentially supported we mean that every
subset of positive Lebesgue measure has positive measure with respect to some spectral
measure. Note that the spectrum may be very far from purely absolutely continuous:
Naboko [8] and Simon [13] constructed potentials with xq(x) → ∞ arbitrarily slowly
for which HD has point spectrum dense in [0, ∞). Our purpose here is to prove
Theorem 1. If q ∈ L2 then the a.c. spectrum of HD is essentially supported by [0, ∞).
In the discrete case, q ∈ `2 implies that hD has a.c. spectrum essentially supported by
[−2, 2].
342 P. Deift, R. Killip

This result is of interest for two reasons in particular: it is optimal in the sense that
the result fails in any Lp (or `p ) space with p > 2, and secondly, the methods used here
are far simpler than those previously employed. That this result is optimal in terms of
Lp -type decay was demonstrated by Kiselev, Last and Simon [7]. Indeed their analysis
of sparse ‘bump’ potentials shows that there exist potentials decaying to zero at infinity
and belonging to ∩2<p≤∞ Lp (or ∩2<p≤∞ `p ) such that HD (or hD ) has purely singular
continuous spectrum.
Using the theory of trace class perturbations [6], Theorem 1 may be extended to:
Corollary 1. If q ∈ L1 +L2 then for all self-adjoint boundary conditions sin(φ)ψ 0 (0)+
cos(φ)ψ(0) = 0, φ ∈ [0, π ) the corresponding half-line Schrödinger operator has a.c.
spectrum essentially supported by [0, ∞).

Corollary 2. If q ∈ L1 (R) + L2 (R) (resp. q ∈ `2 (Z)) then the whole-line Schrödinger

operators

d 2ψ
Hψ = − + q(x)ψ(x) on L2 (R), (3)
dx 2
hψ = ψ(n + 1) + ψ(n − 1) + q(n)ψ(n) on `2 (Z), (4)

have a.c. spectrum of multiplicity two essentially supported by [0, ∞) (resp. [−2, 2]).

The next section introduces the main object of our analysis, the transmission coeffi-
cient, and derives the basic estimates for this quantity. Section 3 uses the Weyl m-function
and its relationship to the transmission coefficient to prove Theorem 1. Of course, by
Weyl’s (relative) compactness theorem, σess (H ) = [0, ∞) and σess (h) = [−2, 2], and
so it is sufficient to limit our considerations to these two intervals respectively.

2. The Transmission Coefficient

Throughout this section we shall assume q to be continuous on R and supported on a
compact subset of [0, ∞). For such q the following are well known:
1. In the continuum case, for every k ∈ C there exists a solution to

−ψ 00 (x) + q(x)ψ(x) = k 2 ψ(x) ∀x ∈ R (5)

such that ψ(x) = eikx for all x to the right of the support of q. Moreover for each x,
ψ(x) and ψ 0 (x) are analytic functions of k.
2. In the discrete case, for every k ∈ C \ {0} there exists a solution to
 
ψ(n + 1) + ψ(n − 1) + q(n)ψ(n) = k + k1 ψ(n) ∀n ∈ Z (6)

such that ψ(n) = k n for all n to the right of the support of q. Again ψ(n) is an
analytic function of k ∈ C \ {0} for each n.
To the left of the support of q, ψ must satisfy the free Schrödinger equation and so
take the form

ψ(x) = aeikx + be−ikx (continuum),

ψ(n) = ak n + bk −n (discrete).
Absolutely Continuous Spectrum of 1D Schrödinger Operators 343

As ψ depends analytically on k so do a and b. Since e±ikx , k ±n represent waves prop-

agating to the right/left in the time dependent picture, it is natural to term t = 1/a and
r = b/a the transmission and reflection coefficients respectively. The proof of Theorem
1 rests on the estimates
Z Z
1 2 π
log k dk ≤ 8 q 2 (x)dx (continuum), (7)
R t
Z π X
1
log sin2 (θ)dθ ≤ π4 q(n)2 (discrete), (8)
−π t n
which are by no means new. They follow from well-known R inverse scattering identities
(10),(13)
P which are true for more general potentials: |q(x)|(1 + |x|)dx < ∞ (or
|q(n)|(1 + |n|) < ∞) is sufficient. See, for example, [4]. These identities are not so
well known in the Schrödinger operator community and so, for the convenience of the
reader, we include proofs under the mild simplification that q is continuous of compact
support.

2.1. Continuum. We begin by rewriting (5) as the integral equation

Z ∞h i
i
v(x) = 1 + 2k 1 − e2ik(y−x) q(y)v(y)dy, (9)
x

where v(x) = e−ikx ψ(x). For Im(k) ≥ 0 and |k| > 0 this may be solved by repeated
substitution. In turn this gives
Z ZZ h i
a = 1 + 2ki q(x)dx − 8k12 1 − e2ik|x−y| q(x)q(y)dxdy + O(k −3 ).

Let γR denote the anticlockwise contour parameterized by k(θ ) = Reiθ , θ ∈ [0, π]. In
light of the above Z Z
π
lim log(a)k 2 dk = − q(x)2 dx.
R→∞ γR 8
Since a may have zeros, log(a) need not be analytic in the upper half-plane. However at
any point Im(k) > 0, where a(k) vanishes, (5) must have an L2 solution. That is a(k) = 0
precisely when k 2 is a negative eigenvalue of the whole line Schrödinger equation (5).
Since q is compactly
R supported there are only finitely many such eigenvalues, indeed no
more than 1 + |xq(x)|dx, refer to [12]. Let {iβm } enumerate the (necessarily simple)
zeros of a and define the corresponding Blaschke product for the upper half-plane
Y k − iβm
B= .
m
k + iβm

So log(Bt) = − log(a/B) is analytic in the upper half-plane. Because a(−k) = a(k)

and B(−k) = B(k) (for k real), Im[log(Bt)] is an odd function of k ∈ R. Thus
Z Z
1 2
log k dk = lim [− log(a) + log(B)]k 2 dk
t R→∞ γR
Z X
π
= 8 q(x)2 dx − 2π3 βm3
, (10)
m
which provides the desired estimate (7).
344 P. Deift, R. Killip

2.2. Discrete. In the discrete case one may rewrite (6) as

∞ h
X i
k
v(n) = 1 − k 2 −1
1 − k 2(n−m) q(n)v(n) (11)
m=n

with v(n) = k n ψ(n). For |k| > 1 one may solve this by repeated substitution, from
which
X XXh i
a = 1 − k1 q(n) + 2k12 1 − k −2|m−n| q(m)q(n) + O(k −3 ). (12)

Just as above a(k) has zeros in {|k| > 1} at precisely those points for which k + 1/k is
an eigenvalue of (6). Let k ∈ {βm } enumerate these zeros. As h is self-adjoint, these βm
must be real. Since q is compactly supported u(n) 7 → q(n)u(n) is a finite rank operator,
hence there only finitely many such βm [6]. We introduce the Blaschke product
Y βm − k βm
B= .
m
1 − βm k |βm |

If k = eiθ then Im[log(Bt)] is an odd function of θ and so, by (12),

Z π Z
1 (k 2 − 1)2
2
π log sin 2
(θ)dθ = 1
log(B/a) dk
−π t 2πi
|k|=1 k3
X X X β 2 − β −2
m m
= 21 q(n)2 + log(βm 2
)− , (13)
n m m
2

where the integral dk is taken anticlockwise. As β 2 − β −2 − log(β 4 ) > 0 for β > 1,

we have the crucial estimate for the discrete case (8).

3. Application to Spectral Theory

We now return our attention to half-line Schrödinger operators: (1) and (2). The central
tool in the spectral analysis of such operators is the Weyl m-function [3,5]. The spectral
significance of the m-function is the representation
Z  
1 E
m(z) = A + − dµ(E) for z 6 ∈ supp(µ) (14)
R E−z 1 + E2
with A ∈ C+ and µ a locally finite measure such that HD (or hD ) is unitarily equivalent
to multiplication by E in L2 (R; dµ). In fact m(z) has a simple expression in terms of
∂2
the Green’s function: m(z) = ∂x∂y G (x, y) in the continuum case and m(z) =
(0,0) z
Gz (0, 0) in the discrete case. As such, it can be recovered from the solutions ψ of (5)
and (6) provided q is of compact support. In the continuum case m(k 2 ) = ψ 0 (0)/ψ(0)
for Im(k) > 0, Re(k) 6 = 0 and m(k + k −1 ) = −ψ(0)/ψ(−1) for |k| > 1, Re(k) 6 = 0
in the discrete case. As mentioned in Sect. 2, ψ depends analytically on k ∈ C (or
k ∈ C \ {0} in the discrete case), so the above definitions meromorphically continue m.
Standard techniques show that (14) implies
weak- lim Im[m(E + i)]dE = π dµ(E). (15)
→0
Absolutely Continuous Spectrum of 1D Schrödinger Operators 345

For q of compact support HD (resp. hD ) has purely absolutely continuous spectrum in

[0, ∞) (resp. [−2, 2]); refer for instance Ch. XIII §8,13 of [10]. Hence (15) becomes
dµ(E) = 1
π Im[m(E + i0)]dE
for E ∈ [0, ∞) or [−2, 2]. By m(E + i0) we mean the limiting value for m(E + i)
as  → 0. Since m has a meromorphic continuation in k this is equivalent to taking the
corresponding value for k.
Thus far we have considered only potentials q which are continuous and of compact
support. To prove Theorem 1 for general L2 (or `2 ) potentials we require two lemmas
regarding limits.
Lemma 1. If qn → q in L2 (or `2 ) then the corresponding Weyl m-functions converge
uniformly on compact subsets of the upper half-plane. And so µn converges weakly to
µ, µn * µ.
Proof. This follows from the relation between the m-function and the Green’s function.
For Im(z) > 0 the second resolvent equation implies that the Green’s function (and the
requisite derivatives) converge uniformly on compact subsets of C \ R. That µn * µ
then follows by the Stone–Weierstrass Theorem. u t
Lemma 2. Let  be an open subset of R and let w(x) ∈ L1loc () be a.e. positive. If µn
is a sequence of positive measures which converge weakly to µ then,
Z  w(K) 
1

lim − log dµ
dx
n 1
w(x) w(x)dx ≥ log (16)
n w(K) K µ(K)
R
for each compact K ⊂  of positive Lebesgue measure. By w(K) we mean K w(x)dx.
Proof. Let K be fixed and let 8 denote the set of positive continuous functions of
compact support which take the value 1 on K. Since µ is regular,
Z Z
µ(K) = inf φdµ = inf lim φdµn
φ∈8 φ∈8 n
Z
≥ inf lim φ dµ dx dx
n
φ∈8 n
Z
dµn 1
≥ lim dx w(x) w(x)dx.
n K
Dividing by w(K) and applying g(t) = − log(t) to both sides, (16) follows from Jensen’s
inequality. Note that because g is decreasing the inequality is reversed and lim becomes
lim. u
t
At last we come to the
Proof of Theorem 1. Let {qn } be a sequence of continuous functions of compact support
which converge to q in L2 . To each such potential we shall associate its m-function
mn , m and corresponding measure µn , µ. From our foregoing remarks, Theorem 1 can
be proved by demonstrating that µ(K) > 0 for all compact sets K ⊆ (0, ∞) (or (−2, 2)
in the discrete case) of positive Lebesgue measure. From Lemmas 1 & 2 we see that it
is sufficient to prove that
Z   Z  
dµn 1 Im mn (E + i0)
− log dE w(E) w(E)dE = − log w(E)dE (17)
K K π w(E)
346 P. Deift, R. Killip

is bounded above (uniformly in n) for some positive weight function. At this point
the proofs for the continuum and discrete cases diverge slightly. We shall treat them
separately.

Continuum. As qn is continuous of compact support, the m-function is just ψ 0 (0)/ψ(0)

and so
an (k) − bn (k) 1 − rn (k)
mn (k 2 ) = ik = ik
an (k) + bn (k) 1 + rn (k)
or conversely rn = (ik − mn )/(ik + mn ). Thus for k 2 ∈ K, k > 0 (which corresponds
to E = k 2 + i0)
4kIm mn 4Im mn
|tn |2 = 1 − |rn |2 = ≤ .
|mn + ik| 2 k
The last inequality follows from the fact that Im(m)√ ≥ 0 for such k; this follows from
(14). Thus choosing weight function w(E) = E/4π we have
Z  
4Immn √
RHS(17) = 4π 1
− log √ EdE
K E
Z
1 2
≤ π1 log k dk
k 2 ∈K,k>0 tn
Z
1 2
≤ π1 log k dk
R tn
Z
≤ 18 q 2 dx,

which proves that (17) is indeed uniformly bounded.

Discrete. As in the continuous case we may write m in terms of r for compactly sup-
ported potentials:
−1 1 + kmn
rn = .
k mn + k
So for k = eiθ , θ ∈ (0, π ), which corresponds to E = 2 cos(θ ) + i0,
4 sin(θ)Im(mn ) 4Im(mn )
|t|2 = 1 − |r|2 = ≤ .
|mn + k|2 sin(θ )
As before the final inequality follows because Im m ≥ 0 for such √ k. Define K̃ = {θ ∈
(−π, π) : 2 cos(θ) ∈ K}. Then with weight function w(E) = 4 − E 2 /8π ,
Z  
4Immn
RHS(17) = 2π 1
− log sin2 (θ )dθ
K̃ sin(θ )
Z
1
≤ 2π1
log sin2 (θ )dθ
K̃ tn
Z π
1
≤ 2π1
log sin2 (θ )dθ
−π tn
X
≤ 21 q2.
Thus (17) is bounded uniformly in n. u
t
Absolutely Continuous Spectrum of 1D Schrödinger Operators 347

Acknowledgements. The authors would like to thank Barry Simon for initiating their collaboration. The work
of the first author was supported in part by NSF Grant #DMS-9500867.

References
1. Christ, M., Kiselev, A., Remling, C.: The absolutely continuous spectrum of one-dimensional Schrödinger
operators with decaying potentials. Math. Res. Lett. 4, no. 5, 719–723 (1997)
2. Christ, M., Kiselev, A.: Absolutely continuous spectrum for one-dimensional Schrödinger operators with
slowly decaying potentials: some optimal results. J. Am. Math. Soc. 11, 771–797 (1998)
3. Coddington, E. A., Levinson, N.: Theory of ordinary differential equations. New York–Toronto–London:
McGraw-Hill Book Company, Inc., 1955, MR 16, 1022b
4. Faddeev, L. D., Takhtajan, L. A.: Hamiltonian methods in the theory of solitons. Translated from the
Russian by A. G. Reı̆man, Berlin: Springer, 1987, MR 89m:58103
5. Gesztesy, F., Simon, B.: m-functions and inverse spectral analysis for finite and semi-infinite Jacobi
matrices. J. Anal. Math. 73, 267–297 (1997)
6. Kato, T.: Perturbation theory for linear operators. Reprint of the 1980 edition, Berlin: Springer, 1995;
MR 96a:47025
7. Kiselev, A., Last, Y., Simon, B.: Modified Prüfer and EFGP transforms and the spectral analysis of
one-dimensional Schrödinger operators. Commun. Math. Phys. 194, 1–45 (1998)
8. Naboko, S. N.: On the dense point spectrum of Schrödinger and Dirac operators. Teoret. Mat. Fiz. 68,
no. 1, 18–28 (1986); MR 88h:81029
9. Reed, M., Simon, B.: Methods of modern mathematical physics. I. Functional analysis. New York–
London: Academic Press, 1972
10. Reed, M., Simon, B.: Methods of modern mathematical physics. IV. Analysis of operators. New York–
London: Academic Press, 1978
11. Remling, C.: The absolutely continuous spectrum of one-dimensional Schrödinger operators with decay-
ing potentials. Commun. Math. Phys. 193, no. 1, 151–170 (1998)
12. Simon, B.: An introduction to the self-adjointness and spectral analysis of Schrödinger operators. Acta
Phys. Aust. 17, 19–42 (1977)
13. Simon, B.: Some Schrödinger operators with dense point spectrum. Proc. Am. Math. Soc. 125, no. 1,
203–208 (1997); MR 97c:34179

Communicated by B. Simon