Time-Dependent SCHR Odinger Equation: Statistics of The Distribution of Gaussian Packets On A Metric Graph

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

ISSN 0081-5438, Proceedings of the Steklov Institute of Mathematics, 2010, Vol. 270, pp. 246–262.


c Pleiades Publishing, Ltd., 2010.
Original Russian Text 
c V.L. Chernyshev, 2010, published in Trudy Matematicheskogo Instituta imeni V.A. Steklova, 2010, Vol. 270,
pp. 249–265.

Time-Dependent Schrödinger Equation:


Statistics of the Distribution of Gaussian Packets
on a Metric Graph
V. L. Chernyshev a,b
Received April 2009

Abstract—We consider a time-dependent Schrödinger equation in which the spatial variable


runs over a metric graph. The boundary conditions at the vertices of the graph imply the
continuity of the function and the zero sum of the one-sided derivatives taken with some weights.
In the semiclassical approximation, we describe a propagation of Gaussian packets on the graph
that are localized at a point at the initial instant of time. The main focus is placed on the
statistics of the behavior of asymptotic solutions as time increases. We show that the calculation
of the number of quantum packets on a graph is related to the well-known number-theoretic
problem of finding the number of integer points in an expanding simplex. We prove that the
number of Gaussian packets on a finite compact graph grows polynomially. Several examples
are considered. In a particular case, Gaussian packets are shown to be distributed on a graph
uniformly with respect to the edge travel times.
DOI: 10.1134/S008154381003020X

1. INTRODUCTION
1.1. Preliminary remarks. In this paper we study the properties of solutions to a time-
dependent Schrödinger equation on a metric graph in the semiclassical approximation.
The theory of equations on such sets is rather young: among the first papers were [2] and [22].
However, this theory has become widely popular in recent years. Differential equations on spatial
networks are used in modeling various problems of natural science: vibrations of elastic networks,
processes in the networks of waveguides, electron states in molecules, etc. Results related to the
analysis of these equations, as well as examples of physical problems that lead to equations on
graphs and results on the relation between these problems and the theory of partial differential
equations, are collected, e.g, in survey [20] and book [9]. In particular, in [17] the authors discuss
the problem of convergence of the spectrum of a family of operators on a graphlike thin manifold
to the spectrum of an operator on a graph.
Note that the questions discussed in the present paper are close to some boundary value problems
for hyperbolic equations on networks. Interesting results in this field were obtained in [1, 3, 4, 6,
11, 13, 15].
The semiclassical approximation is discussed, for example, in book [8].

1.2. Basic definitions. A metric graph Γ is a one-dimensional cell complex. In other words,
not only vertices but also the interior points of edges are included in this set. We denote the vertices
by aj and the edges by γj . We consider only graphs with a finite number of vertices and edges. Loops
a Bauman Moscow State Technical University, Vtoraya Baumanskaya ul. 5, Moscow, 105005 Russia.
b Moscow State University, Moscow, 119991 Russia.

E-mail address: vchern@mech.math.msu.su

246
TIME-DEPENDENT SCHRÖDINGER EQUATION 247

and multiple edges are not excluded. By Γ(a) we denote the set of edges incident to the vertex a.
Denote the numbers of vertices, edges, and pendant vertices by M , N , and K, respectively.
Define a differential operator H  on a graph as follows (see [9]).
Each edge is identified with a segment of the real axis. We introduce the space L2 (Γ) of functions
on a graph,
N
2
L (Γ) = L2 (γj ),
j=1

with the standard inner product



(f, g) = f (x)g(x) dx. (1)
Γ
Let V be an arbitrary real-valued continuous function on Γ that is smooth on the edges. Then
the Schrödinger operator
 d2 ψ(x)
Hψ(x) = −h2 + V (x)ψ(x) (2)
dx2

is defined on the set of functions that belong to the Sobolev space, ψ ∈ N 2
j=1 W2 (γj ), and satisfy
the following boundary conditions at the vertices:
(i) the function ψ is continuous on Γ;
(ii) the transmission conditions
 dψj
αj (am ) = 0, αj ∈ R, m = 1, 2, . . . , M − K, (3)
dx
γj ∈Γ(am )

hold at all internal vertices (i.e., at vertices of valency greater than 1);
(iii) ψ(am ) = 0 at all external (pendant) vertices, i.e., at vertices of valency 1.
We say that the transmission conditions have the form of Kirchhoff conditions if all coefficients
in the transmission conditions have the plus sign for outgoing edges and the minus sign for incoming
edges at each vertex. If, in addition, the coefficients at each vertex have the same absolute value,
then these conditions are said to be natural.
 is self-adjoint when
In [2] (see a similar proposition in [9]), it was proved that the operator H
the coefficients of the second derivative in the equations on the edges are equal to the corresponding
coefficients in the transmission conditions. In our case, all the coefficients of the second derivatives
are equal to each other (and equal to −h2 ). Thus, the operator is certainly self-adjoint if the
transmission conditions are natural (see also [12]).
Consider the time-dependent Schrödinger equation
∂ψ(x, t) ∂ 2 ψ(x, t)
ih = −h2 + V (x)ψ(x, t), (4)
∂t ∂x2
where V is a function smooth on the edges (a potential). The corresponding Hamiltonian is given
by H = p2 + V (x, t) (see, for example, [8]).
We take the initial conditions (for the Cauchy problem) in the form of a narrow packet (a Gaus-
sian packet) localized as h → 0 near a point x0 :
 
i(a(x − x0 )2 + b(x − x0 ) + c)
ψ(x, 0) = K exp . (5)
h
Here b and c are real constants, and the imaginary part of a is greater than zero. The coefficient K
has the form K = h−1/4 K1 , K1 ∈ R. We introduced the normalizing factor h−1/4 to ensure that
ψ(x, 0) = O(1) in the norm of the space L2 .

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


248 V.L. CHERNYSHEV

2. PROPAGATION OF QUANTUM PACKETS ON A METRIC GRAPH


2.1. Maslov’s complex germ on a straight line. First, recall the well-known procedure
for constructing semiclassical Gaussian packets on a straight line.
Proposition 2.1 (see [7]). The function
⎛ ⎞
t  
pl (t) ⎠ i S0 (t) + S1 (t)(x − X(t)) + S2 (t)(x − X(t))2
ψ(x, t) = K exp⎝− dt exp (6)
ql (t) h
0

is a solution to the Cauchy problem for the equation

 ∂ψ(x, t)
Hψ(x, t) = ih + O(h3/2 ).
∂t

If the operator H  on the left-hand side is self-adjoint, then this solution differs from the exact
solution of the time-dependent Schrödinger equation by at most O(h1/2 ). Here

t

S0 (t) = c + (P (t))2 − V0 (t) dt, (7)
0
t t
S1 (t) = b − V1 (t) dt = b + P (t) dt, (8)
0 0

pl (t)
S2 (t) = ; (9)
2ql (t)

X(t), P (t) is a solution of the Hamiltonian system

Ẋ(t) = Hp (X(t), P (t)),


(10)
Ṗ (t) = −Hx (X(t), P (t))

with the initial data X(0) = x0 and P (0) = b; and ql (t), pl (t) is a solution of the linearized system

q̇l = 2pl ,
(11)
∂2V
ṗl = − ql .
∂x2
One can take the following initial data for this system: ql (0) = 1 and pl (0) = 2a.
In what follows, we assume that there are no turning points on the graph (see [7]); such points,
if they exist, can be treated in a standard way.
2.2. The case of a single pendant vertex. We begin with the simplest example: suppose
that the graph consists of a single vertex O and a single edge (of infinite length). We will need the
results of this subsection below when considering an arbitrary graph.
Assume that the vertex corresponds to x = 0 and impose the Dirichlet condition ψ(0, t) = 0 at
this point.
The configuration space is a half-plane (R \ R+ for the coordinate x and R for t). Let p and E
be the respective conjugate coordinates.

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


TIME-DEPENDENT SCHRÖDINGER EQUATION 249

We will seek a solution as a sum of incident and reflected waves:


 (1)   (2) 
iS (x, t) (1) iS (x, t) (2)
ψ(x, t) = exp φ0 (x, t) + exp φ0 (x, t) + O(h1/2 ). (12)
h h
Repeating the calculations from [7], we obtain the Hamilton–Jacobi and transport equations for
each of the functions. Solutions of these equations can be written out in a standard way. To find
S (1) and φ(1) , we again set initial conditions, and to find S (2) and φ(2) , we set boundary conditions.
The Hamiltonian system is supplemented with two equations, ṫ = 1 and Ė = 0.
The initial conditions are X (1) (0) = x0 , P (1) (0) = b, t(1) = τ (and τ = 0), E (1) = −H(x0 , b),
(1) (1) (1) (1)
ql (0) = 1, pl (0) = 2a, S0 (0) = c, S1 (0) = b, and C = K.
Accordingly, the boundary conditions are X (2) (β) = 0, P (2) (β) = −P (1) (β), t(2) = τ + β
(2) (1) (2) (1) (2) (1)
(and τ = 0), E (2) = −H(0, −P (1) (β)), ql (β) = ql (β), pl (β) = pl (β), S0 (β) = S0 (β),
(2) (1)
S1 (β) = −S1 (β), and C = K. Here β is the instant of reflection, i.e., the time taken for
a Gaussian packet to reach the vertex of the half-line. Representing the function ψ in (6) as
φ(1) exp(iS (1)/h), one can find the description of the functions S (1) and φ(1) in Proposition 2.1.
Now, let us describe S (2) and φ(2) .
Proposition 2.2 (cf. [24]). Expression (12) is a solution to the Cauchy problem for the equa-
tion
 ∂ψ(x, t)
Hψ(x, t) = ih + O(h3/2 ).
∂t
If the operator H  on the left-hand side is self-adjoint, then this solution differs from the exact
solution of the time-dependent Schrödinger equation by at most O(h1/2 ).
The first term of (12) is described in Proposition 2.1, and the second is
⎛ β ⎞
 (1) t (2)
pl (t) pl (t)
ψ(x, t) = K exp⎝− (1)
dt − (2)
dt⎠
ql (t) ql (t)
0 β
 (2) (2) (2) 
i S0 (t) + S1 (t)(x − X(t)) + S2 (t)(x − X(t))2
× exp . (13)
h
Here
t
(2) (1) 
S0 (t) = S0 (β) + (P (2) (t))2 − V0 (t) dt, (14)
β

t t
(2) (1)
S1 (t) = −S1 (β) − V1 (t) dt = b + P (2) (t) dt, (15)
β β
(2)
(2) pl (t)
S2 (t) = (2)
; (16)
2ql (t)
X (2) (t), P (2) (t) is a solution to the Hamiltonian system

Ẋ (2) (t) = Hp X (2) (t), P (2) (t) ,

Ṗ (2) (t) = −Hx X (2) (t), P (2) (t) ,
(17)
ṫ(2) = 1,

Ė (2) = 0

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


250 V.L. CHERNYSHEV

with the boundary conditions X (2) (β) = 0, P (2) (β) = −P (1) (β), t(2) = τ + β, and E (2) =
(2) (2)
−H(0, −P (1) (β)); and ql (t), pl (t) is a solution to the linearized system (11). As the initial
(2) (1) (2) (1)
data for this system, one can take ql (β) = ql (β), pl (β) = pl (β), and
⎛ ⎞
β (2)
pl (t)
C = K exp⎝− (2)
dt⎠ .
ql (t)
0

Proof. Formulas for the solutions follow from Proposition 2.1. In addition, we should verify
that the second term is exponentially small for 0 < t < β and the first term is exponentially small
for t > β.
On the trajectory, we have P 2 (t) + V (X(t)) = E. Hence, in the absence of turning points, P (t)
may only be either negative or positive. By virtue of the Hamilton equations, we have Ẋ(t) > 0 for
the first term in (12) and Ẋ(t) < 0 for the second. Let us apply Lagrange’s mean value theorem
to the function X(t) on the interval from arbitrary t to t = β: X(t) = Ẋ(ξ)(t − β). Hence we find
that X (1) (t) > 0 for t > β and X (2) (t) > 0 for t < β. Since the dissipativity condition (Im S > 0) is
preserved under a Hamiltonian flow (see [7]), we find that until t = β the asymptotic solution is
defined by the first term, while after t = β, by the second (recall that the configuration space
is defined by the inequality x ≤ 0).
This proves the well-posedness of the problem with the initial conditions set only for the first
term.
The fact that ql (t) does not vanish follows from the dissipativity of the germ (i.e., from the fact
that Im S2 preserves its sign), which holds because the linearized system is Hamiltonian. Let us
prove this:
 
1 pl (t) 1 1
Im S2 (t) = Im = Im pl q l = (pl ql − ql pl ). (18)
2 ql (t) |ql (t)|2 4i|ql (t)|2

Hence we find that the sign of Im S2 (t) is determined by the expression ql , pl  = pl ql −ql pl . However,
this quantity is independent of t. To show this, one should take the derivative of ql , pl  and verify
that it vanishes along trajectories of the linearized Hamiltonian system. Thus, this quantity does
not change with t, and at the initial instant it is different from zero because Im S2 (0) = Im a > 0.
Now, if we assume that ql = 0, we obtain ql , pl  = 0, which contradicts the dissipativity.
2.3. Two infinite rays meeting at a single point. Consider two infinite rays meeting at a
single point (see Fig. 1). On the edge on which the initial data are specified, we will seek a solution
in the form
 (1)   (2) 
iS (x, t) (1) iS (x, t) (2)
u(γ1 ) (x, t) = exp φ0 (t) + exp φ0 (t) + O(h1/2 ). (19)
h h

On the other edge, we seek a solution in the form


 (3) 
(γ2 ) iS (x, t) (3)
ψ(x, t) = u (x, t) = exp φ0 (t) + O(h1/2 ). (20)
h

Here u(γj ) (x, t) stands for the solution restricted to the jth edge.
(k,j)  (k,j)
Denote ψ (k,j) (x, t) = exp iS h (x,t) φ0 (t), where k is the number of the edge and j is the
number of a quantum packet on the edge.
Let us supplement the Schrödinger equation with the conditions of continuity of the solution
on the entire graph, u(γ1 ) (0, t) = u(γ2 ) (0, t), and with the transmission conditions, which in general

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


TIME-DEPENDENT SCHRÖDINGER EQUATION 251

α1

a α2
Fig. 1. Two infinite rays and one vertex.

have the form


 ∂u(γj )
αj (0, β) = 0. (21)
∂x
γj ∈Γ(a)

In our situation these conditions reduce to

∂u(γ1 ) ∂u(γ2 )
α1 (0, β) + α2 (0, β) = 0. (22)
∂x ∂x

Using our notation, we obtain u(γ1 ) (x, t) = ψ (1,1) (x, t) + ψ (1,2) (x, t) and u(γ2 ) (x, t) = ψ (2,1) (x, t).
The conditions take the form

ψ (1,1) (x, t) + ψ (1,2) (x, t) = ψ (2,1) (x, t), (23)


 (1,1) 
∂ψ ∂ψ (1,2) ∂ψ (2,1)
α1 (0, β) + (0, β) + α2 (0, β) = 0. (24)
∂x ∂x ∂x

Let us take into account the fact that

∂S (1,2) ∂S (1,1) ∂S (2,1) ∂S (1,1)


(0, β) = − (0, β), (0, β) = − (0, β).
∂x ∂x ∂x ∂x
(1,1) (1,2) (2,1)
It follows from (23) that S0 (0, β) = S0 (0, β) = S0 (0, β). In addition, the following
(1,1) (1,2) (2,1)
equality holds: φ0 (β) + φ0 (β) = φ0 (β). Then, from (24) we obtain

∂S (1,1) (1,1) (1,2) (2,1) 


(0, β) α1 φ0 (β) − α1 φ0 (β) − α2 φ0 (β) = 0. (25)
∂x
∂S (1,1)
The absence of turning points implies that ∂x = 0. Hence,

(1,1) (1,2) (1,1) (1,2)


α1 φ0 (β) − α1 φ0 (β) − α2 (φ0 (β) + φ0 (β)) = 0. (26)

As a result, we obtain
(1,2) α1 − α2 (1,1)
φ0 (β) = φ (β) (27)
α1 + α2 0
for the amplitude of the reflected quantum packet and

(2,1) 2α1 (1,1)


φ0 (β) = φ (β) (28)
α1 + α2 0
for the amplitude of the transmitted quantum packet.
(1,1)  (1,1)
The initial data completely determine ψ (1,1) (x, t) = exp iS h (x,t) φ0 (x, t).
Thus, we have proved the following proposition.

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


252 V.L. CHERNYSHEV

Proposition 2.3 (cf. [24]). On a metric graph consisting of two infinite edges connected at
a single vertex, the leading term of the asymptotic solution to the Cauchy problem for the time-
dependent Schrödinger equation with the initial conditions (5), continuity conditions, and the trans-
mission conditions (22) is given by (19) and (20).
The term ψ (1,1) is determined by the initial conditions.
The term ψ (1,2) is given by
⎛ β ⎞
 (1,1) t (1,2)
α2 − α1 pl (t) pl (t)
ψ (1,2) (x, t) = K exp⎝− (1,1)
dt − (1,2)
dt⎠
α1 + α2 q (t) q (t)
0 l β l

 (1,2) (1,2) (1,2) 


i S0 (t) + S1 (t)(x − X(t)) + S2 (t)(x − X(t))2
× exp . (29)
h
Here
t
(1,2) (1,1) 
S0 (t) = S0 (β) + (P (1,2) (t))2 − V0 (t) dt, (30)
β

t t
(1,2) (1,1)
S1 (t) = −S1 (β) − V1 (t) dt = b + P (1,2) (t) dt, (31)
β β
(1,2)
(1,2) pl (t)
S2 (t) = (1,2)
; (32)
2ql (t)

X (1,2) (t), P (1,2) (t) is a solution to the Hamiltonian system



Ẋ (1,2) (t) = Hp X (1,2) (t), P (1,2) (t) ,

Ṗ (1,2) (t) = −Hx X (1,2) (t), P (1,2) (t) ,
(33)
ṫ(1,2) = 1,

Ė (1,2) = 0

with the boundary conditions X (1,2) (β) = 0, P (1,2) (β) = −P (1,1) (β), t(1,2) = τ + β, and E (1,2) =
(1,2) (1,2)
−H(0, −P (1,2) (β)); and ql (t), pl (t) is a solution of the linearized system (11). As the initial
(1,2) (1,1) (1,2) (1,1)
data for this system, one can take ql (β) = ql (β) and pl (β) = pl (β).
The term ψ (2,1) (the “transmitted” quantum packet) has the form
⎛ β ⎞
 (1,1) t (2,1)
2α p (t) p (t)
K exp⎝− dt⎠
2
ψ (2,1) (x, t) = l
(1,1)
dt − l
(2,1)
α1 + α2 q (t) q (t)
0 l β l

 (2,1) (2,1) (2,1) 


i S0 (t) + S1 (t)(x − X(t)) + S2 (t)(x − X(t))2
× exp . (34)
h
Here
t
(2,1) (1,1) 
S0 (t) = S0 (β) + (P (2,1) (t))2 − V0 (t) dt, (35)
β

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


TIME-DEPENDENT SCHRÖDINGER EQUATION 253

t t
(2,1) (1,1)
S1 (t) = −S1 (β) − V1 (t) dt = b + P (2,1) (t) dt, (36)
β β
(2,1)
(2,1) pl (t)
S2 (t) = (2,1)
; (37)
2ql (t)

X (2,1)(t), P (2,1)(t) is a solution to the Hamiltonian system with the boundary conditions X (2,1)(β) = 0,
(2,1) (2,1)
P (2,1) (β) = −P (1,1) (β), t(2,1) = τ + β, and E (2,1) = −H(0, −P (2,1) (β)); and ql (t), pl (t) is a
(2,1)
solution of the linearized system (11). As the initial data for this system, one can take ql (β) =
(1,1) (2,1) (1,1)
ql (β) and pl (β) = pl (β).
Remark. The amplitudes of the transmitted and reflected packets are determined by the
coefficients α1 and α2 in the transmission conditions. Note that the Schrödinger operator is self-
adjoint if α1 = α2 ; in this case, the reflection coefficient is zero, and the packet passes from one
edge of the graph to the other “without feeling” the vertex.
2.4. Three infinite rays meeting at a single point. Following the same line as in the
previous case (and using the formulas obtained there), consider a graph obtained by joining three
rays at a single point.
On the edge on which the initial data are specified, we will seek a solution in the form
 (1,1)   (1,2) 
iS (x, t) (1,1) iS (x, t) (1,2)
ψ(x, t) = exp φ0 (t) + exp φ0 (t) + O(h1/2 ). (38)
h h

On the other two edges, we seek a solution in the form


 (k,1) 
iS (x, t) (k,1)
ψ(x, t) = exp φ0 (t) + O(h1/2 ), k = 2, 3. (39)
h
(1,1)  (1,1)
The initial data completely determine ψ (1,1) = exp iS h (x,t) φ0 (t).
Using the notation introduced in Subsection 2.3,

u(γ1 ) (x, t) = ψ (1,1) (x, t) + ψ (1,2) (x, t),

u(γ2 ) (x, t) = ψ (2,1) (x, t),

u(γ3 ) (x, t) = ψ (3,1) (x, t),

we obtain the conditions

ψ (1,1) (x, t) + ψ (1,2) (x, t) = ψ (2,1) (x, t) = ψ (3,1) (x, t), (40)
 (1,1) 
∂ψ ∂ψ (1,2) ∂ψ (2,1) ∂ψ (3,1)
α1 (0, β) + (0, β) + α2 (0, β) + α3 (0, β) = 0. (41)
∂x ∂x ∂x ∂x

Let us take into account the fact that

∂S (1,2) ∂S (1,1) ∂S (2,1) ∂S (1,1)


(0, β) = − (0, β), (0, β) = − (0, β),
∂x ∂x ∂x ∂x
∂S (3,1) ∂S (1,1)
(0, β) = − (0, β).
∂x ∂x
PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010
254 V.L. CHERNYSHEV
(1,1) (1,2) (2,1) (3,1)
It follows from (40) that S0 (0, β) = S0 (0, β) = S0 (0, β) = S0 (0, β). Moreover,
(1,1) (1,2) (2,1) (3,1)
φ0 (β) + φ0 (β) = φ0 (β) = φ0 (β). Then, from (41) we derive
(1,1) (1,2) (1,1) (1,2)  (1,1) (1,2) 
α1 φ0 (β) − α1 φ0 (β) − α2 φ0 (β) + φ0 (β) − α3 φ0 (β) + φ0 (β) = 0. (42)

As a result, we obtain the expression

(1,2) α1 − α2 − α3 (1,1)
φ0 (β) = φ (β) (43)
α1 + α2 + α3 0
for the amplitude of the reflected quantum packet and the identical expressions

(2,1) (3,1) 2α1 (1,1)


φ0 (β) = φ0 (β) = φ (β) (44)
α1 + α2 + α3 0
for the amplitudes of the transmitted quantum packets.
2.5. Propagation of quantum packets on an arbitrary graph. The arguments given
above allow us to describe the propagation of Gaussian packets on an arbitrary metric graph.
On each edge, this is done by means of Maslov’s complex germ (i.e., by finding solutions to the
Hamiltonian system and its linearization). To describe the behavior of Gaussian packets at the
vertices of the graph, it suffices to consider the case of a star graph.
Generalizing the above arguments, we can formulate the following theorem.
Theorem 2.1. Suppose given a star graph with a single vertex a of valency n. Suppose that
the initial data are given by (5), with the point x0 lying on one of the edges. Then, at any finite
instant of time, the solution to the Cauchy problem for the equation

 ∂ψ(x, t)
Hψ(x, t) = ih + O(h3/2 )
∂t
iS (j) (x,t) 
is a sum of a finite number of quantum packets, i.e., functions of the form exp h φj (t), where
(j) (j) (j) (j)
S (j) = S0 (t) + (x − xj (t))S1 + S2 (x − xj (t))2 , with Im S2 > 0.
More precisely, a packet reaching the vertex a is split into n packets that travel along the edges
incident to this vertex. On each edge, the solution is defined by the Hamiltonian system of ordinary
differential equations (10) and its linearization (11). The initial values of the amplitudes are given
by the following expressions:

(1,2) α1 − α2 − . . . − αn (1,1)
φ0 (β) = φ (β) (45)
α1 + α2 + . . . + αn 0
for the reflected quantum packet, and

(k,1) 2α1 (1,1)


φ0 (β) = φ (β), where k = 2, . . . , n, (46)
α1 + α2 + . . . + αn 0
for the transmitted quantum packets. Here β is the instant of time when the initial packet arrives
at the point a.
 on the left-hand side is self-adjoint, then this solution differs from the exact
If the operator H
solution of the time-dependent Schrödinger equation by at most O(h1/2 ).
Proof. The proof is based on the fact that the form of the solution on each edge is known. We
will assume that the edge along which the packet has arrived at the vertex has number 1. Using
(k,j)  (k,j)
the notation introduced above for an individual packet ψ (k,j) (x, t) = exp iS h (x,t) φ0 (t), we

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


TIME-DEPENDENT SCHRÖDINGER EQUATION 255

write the solutions on the edges as u(γ1 ) (x, t) = ψ (1,1) (x, t) + ψ (1,2) (x, t) and u(γj ) (x, t) = ψ (j,1) (x, t)
for j > 1.
(1,1)  (1,1)
The initial data completely determine ψ (1,1) = exp iS h (x,t) φ0 (t); see Proposition 2.1 for
details.
To set boundary conditions for ψ (1,2) (x, t) and ψ (j,1) (x, t), j > 1, we write out the conditions of
continuity of the solution on the entire graph, u(γ1 ) (0, β) = u(γj ) (0, β), j > 1, and the transmission
conditions
 ∂u(γj )
αj (0, β) = 0. (47)
∂x
γj ∈Γ(a)

Let us take into account that

∂S (1,2) ∂S (1,1)
(0, β) = − (0, β)
∂x ∂x
and
∂S (j,1) ∂S (1,1)
(0, β) = − (0, β)
∂x ∂x
for the indices j > 1.
(1,1) (1,2) (j,1)
It follows from the continuity condition that S0 (0, β) = S0 (0, β) = S0 (0, β), j > 1. In
(1,1) (1,2) (j,1)
addition, φ0 (β) + φ0 (β) = φ0 (β), j > 1.
Combined with the requirement of absence of turning points, these equations lead to the desired
formulas. The well-posedness of the boundary value problem for ψ (1,2) (x, t) is discussed in detail in
the proof of Proposition 2.2.
Thus, we can describe the evolution of a Gaussian packet in the case of an arbitrary graph. At
each vertex reached by a quantum packet, we can apply (45) and (46). We obtain a solution to
equation (4) on a subgraph (of the original graph) that contains a point at which the initial data
are defined. Thus, at any finite instant of time, a semiclassical solution to the Cauchy problem
represents a finite number of Gaussian packets traveling on a metric graph.
Remark. The case of α1 + α2 + . . . + αn = 0 corresponds to a non-self-adjoint operator and has
no physical meaning. In [5] it was proved that the spectrum of the corresponding operator under
some additional conditions can fill the whole complex plane.
One should note the similarity between expressions (45) and (46) and the formulas obtained for
a hyperbolic equation in [10].

3. STATISTICS OF PROPAGATION OF QUANTUM PACKETS


As shown in the previous section, a semiclassical solution to the Cauchy problem for the time-
dependent Schrödinger equation with initial conditions (5) has the form Ψ(x, t) + O(h1/2 ), where
Ψ(x, t) is a finite sum of Gaussian packets. In this section, we consider an asymptotic behavior of
the function Ψ(x, t) as t → ∞; namely, we study how the number of quantum packets varies in
time. Note that this problem differs from the problem of describing the asymptotic behavior of the
solution to the Schrödinger equation as t → ∞, since the estimate for the residual is valid only
for finite times. From the physical point of view, this means that we deal with large times which,
however, are much less than 1/h.
In this section, we consider finite graphs with compact edges. Suppose that the transmission
conditions at the vertices are such that the Schrödinger operator is self-adjoint (see [12]). In this
case, formulas (45) and (46) imply that the number of quantum packets is not changed at vertices

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


256 V.L. CHERNYSHEV

of degree 2 (since the reflected quantum packet has zero amplitude). Following [21], we will call
graphs without degree 2 vertices clean graphs. Below in the text we consider only such graphs.
Moreover, formulas (45) and (46) imply that if a quantum packet passes through a vertex of
degree v, then exactly v new quantum packets are generated.
Consider tj , the travel time of a quantum packet along the jth edge. Recall that, as shown
above, the travel time is determined by the solutions of the Hamiltonian system subject to given
initial conditions. We will assume that tj are linearly independent over the field Q (the situation of
general position).
We will need some number-theoretic propositions on the evaluation of the number of points
with integer coordinates that fall into an expanding polyhedron. The results in this field strongly
depend on whether the coordinates of the vertices of a polyhedron are rational. In the rational case
the results are related to the Ehrhart polynomials (and quasipolynomials) (see [16]) and generalize
Pick’s theorem. Research in this field is currently being intensively developed (see [14] and references
therein). The results concerning the case of irrational coordinates go back to the paper by Hardy
and Littlewood [18], where the case of a right triangle on the plane was analyzed. References to
state-of-the-art research in this field can be found in [23] and [19].
3.1. Asymptotic behavior of the number of quantum packets. Define a function N (T )
as the number of quantum packets on a graph at time instant T .
Let ωj be the frequency of traveling along the jth edge, i.e., ωj = 1/tj .
Proposition 3.1 (cf. [24]). The following formula is valid for a star graph (consisting of a
single vertex of valency v and v degree 1 vertices that are joined to the first vertex ):


v−1  
N (T ) = (v − k) W k (ω1 T /2, . . . , ωk T /2) ; (48)
k=1 1≤i1 <i2 <...<ik ≤v

here W ( k (ω1 T /2, . . . , ωk T /2)) is the number of points of the integer lattice that belong to the
k-dimensional simplex k (ω1 T /2, . . . , ωk T /2) with vertices ω1 T /2, . . . , ωk T /2 on the coordinate
axes. The points belonging to lower dimension simplices of the same type (to the faces of k ) are
not counted.
Writing out the leading term of the asymptotic expression, we obtain
1 t1 + . . . + tv v−1
N (T ) = T + o(T v−1 ). (49)
2v−1 (v − 1)! t1 . . . tv

Proof. For simplicity, we will consider a graph that contains only three edges (v = 3). The
calculations in the general case are similar.
An increase in the number of quantum packets is possible only at the vertices of a graph.
According to the results described in Section 2, at vertices of degree 1 (where the Dirichlet condition
is specified), the number of quantum packets cannot increase. It remains to find out what can occur
at a vertex of degree 3.
By the Cauchy theorem (on the uniqueness of a solution to the differential equation on each
edge), at every fixed instant of time, more than three quantum packets cannot leave (or reach)
a vertex. Thus, the number of quantum packets may either (a) increase by 2 (when one packet
arrives), (b) increase by 1 (when two packets arrive along two different edges), or (c) remain the
same (when quantum packets arrive along all the three edges).
Variant (a) occurs at time instants that are multiples of the doubled  travel
 time
  along
 each
edge; i.e., the total contribution to the number of quantum packets is 2 2tT1 + 2tT2 + 2tT3 . Since

[x] = x − {x} and the second term here is o(x), we obtain tT1 + tT2 + tT3 + o(T ).

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


TIME-DEPENDENT SCHRÖDINGER EQUATION 257

Now, consider variant (b). Three variants are possible for the choice of two out of three edges.
Take, for example, the first and the second edges.
With quantum packets on a graph, we associate rows of the form (i1 , i2 , . . . , im ), where im may
be equal to 1, 2, or 3. Namely, if a quantum packet has first traveled forth and back along the
first edge, then the first place in the row is occupied by 1; if a packet has first traveled along the
second edge, then the first place is occupied by 2; etc. It is clear that, at any instant of time, any
quantum packet can be described by such a row. In addition, if we omit the initial period of time
(when a quantum packet has not yet arrived from the initial point at the first vertex), each row
will correspond to a quantum packet (for a given graph). Note that a row does not describe what
packets (interacting according to the rules described in Section 2) have led to the formation of a
given quantum packet.
To describe situation (b), it suffices to consider only the rows composed of ones and twos.
Indeed, if a packet has arrived at a vertex (for definiteness, along the first edge), then this packet is
described by a row of the form (i1 , i2 , . . . , 1). If a quantum packet has traveled along all the three
edges by the time T , then there are two other packets that arrive at the same vertex along the
other two edges at the same instant of time (and the number of quantum packets is not changed).
Indeed, there exist two other rows that have the same sum of components but contain either 2 or 3
at the last position.
Let us show that the number of quantum packets increases at the time instants corresponding
to points with positive integer coordinates that fall into a right triangle with the sides 2tT1 and 2tT2 .
We must calculate the number of simultaneous arrivals of two quantum packets at a given vertex
(denote the set of such events by A). Denote by B the subset of the set of rows that corresponds
to the events of interest and then calculate the number of elements in this subset.
The simultaneous arrival at a vertex of two quantum packets along different edges can be de-
scribed by two rows with the same sum of components but with different last coordinates. Let us
assign only those rows to this event that contain 2 in the last position. Such rows always exist
because the absence of 2’s among the components implies that a quantum packet has not traveled
along the second edge, which is impossible in situation (b). It is not important how the other
coordinates are arranged (for a given sum), and we will consider the rows in which all 1’s precede
all 2’s.
Let us explain why the correspondence between the elements of the sets A and B is one-to-one.
On the one hand, any event corresponds to at most one row of the type described. Indeed, otherwise
we would obtain two different representations of the form T = 2nt1 + 2mt2 for a given T (using such
a representation, one can easily write out a row of the necessary form); however, this is impossible
because t1 and t2 are linearly independent over Q. We have proved the injectivity. The surjectivity
follows from the fact that each row from the set B corresponds to a time instant T (calculated by
the above formula) at which two quantum packets that have traveled along different edges meet at
the vertex.
We have obtained rows of the form (1, . . . , 1, 2, . . . , 2). They are in one-to-one correspondence
with linear combinations of the form 2nt1 + 2mt2 , where n and m are positive integers. If a time
constraint T is given, we obtain the condition 2nt1 + 2mt2 ≤ T . The solutions of such an inequality
correspond to points with positive integer coordinates that fall into a right triangle with the sides
T T 2
2t1 and 2t2 . The number of points in the triangle is equal to its area up to o(T ), i.e., to the
2
expression 12 4tT1 t2 (in the general case, we deal with the volume of a k-dimensional simplex).
It remains to sum up the number of variants for the choice of two edges and obtain a symmetric
polynomial of degree 2 in frequencies, multiplied by T 2/8.
We need only the leading term of the asymptotic expression; therefore, we take into account
only the term of order T 2 .

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


258 V.L. CHERNYSHEV

t1

t3

t2

Fig. 2. Three edges meet at a single point.

Remark. In the case of three edges (see Fig. 2), using the results presented in [18], we can
write a sharper formula (that includes an explicit expression for the second term of the asymptotic
relation), namely,
1 1
N (T ) = σ2 (ω1 , ω2 , ω3 )T 2 + σ1 (ω1 , ω2 , ω3 )T + o(T ), (50)
8 2
or, passing from ωj to tj ,
 
1 (t1 + t2 + t3 ) 2 1 1 1 1
N (T ) = T + + + T + o(T ). (51)
8 t1 t2 t3 2 t1 t2 t3

Remark. The coefficient of the leading term may be different from the sum of the coefficients
corresponding to the highest degree vertices regarded as vertices of a star graph. For example, the
following proposition is valid.
Proposition 3.2 (cf. [24]). The following formulas hold for a graph consisting of two vertices
connected by v edges (see Fig. 3):


v−1  
N (T ) = (v − k) W k (ω1 T, . . . , ωk T ) , (52)
k=1 1≤i1 <i2 <...<ik ≤v

1 t1 + . . . + tv v−1
N (T ) = T + o(T v−1 ). (53)
(v − 1)! t1 . . . tv

The proof largely repeats the proof of Proposition 3.1. The difference is that an increase in the
number of quantum packets occurs not only when the chosen quantum packet returns to the central
vertex (having traveled along an edge twice) but also when it reaches the other vertex after passing
an edge once. Thus, to calculate the leading term of the asymptotic expression, we have to take
account of how many points with integer positive coordinates satisfy the inequality xt1 + yt2 < T
(this is a “contribution” of the first two edges; then one should consider the other two variants).
The asymptotic behavior of the number of packets in an arbitrary graph is described by the
following theorem.
Theorem 3.1. Let Γ be a compact graph without degree 2 vertices. Let, in addition, the num-
bers t1 , . . . , tM be incommensurable. Then, as T increases, the function N (T ) can be represented as

N (T ) = CT M −1 + o(T M −1 ), (54)

where C is a positive constant and M is the number of edges in the graph.


Proof. The number of packets (or trajectories) can only be changed at the vertices of the
graph. Consider a fixed vertex. Let its valency be v. If k packets simultaneously arrive at the

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


TIME-DEPENDENT SCHRÖDINGER EQUATION 259

vertex, then v − k new packets appear at this instant. The number of such events is described by
the number of nonnegative integers nj that satisfy inequalities of the form

n1 tl1 + . . . + nm tlm ≤ T, (55)

where tj is the travel time of the trajectory along the jth edge. Indeed, packets cannot appear at
our vertex at time instants that are not linear combinations (with positive integer coefficients) of
the edge travel times in the graph. Thus, for every packet arriving at the fixed vertex, we take into
account along which edges the trajectories that generate the packet have passed. The numbers nj
show how many times a packet occurred on the edge with travel time tlj . Since all tlj are linearly
independent over Q, the leading coefficient in the asymptotic expression for the number of packets
(as T increases) is determined by the volume of the simplex defined by inequality (55). Thus, for
large T , the most frequent events are those for which the number of terms on the left-hand side
of inequality (55) is maximal (this follows from the fact that the number of integer points in an
expanding simplex of dimension k grows as T k ). In other words, the number of different edges along
which the Gaussian packets arriving at our vertex have already passed must be maximal. In order
that v − k be nonzero, we must exclude at least one edge. Thus, if the graph contains a vertex
such that one of the edges incident to it can be removed without violating the connectedness of the
graph, then the number of terms on the left-hand side of inequality (55) is maximal and equal to
M − 1. If there are several such vertices, then the number of possible occurrences of new packets is
defined by similar inequalities (i.e., the number of integer points in the (M − 1)-dimensional simplex
is multiplied by a number that depends only on the graph). In a connected graph containing more
than one edge, there always exits a vertex of the necessary type. The rejection of packets whose
birth is determined by inequalities with a smaller number of terms on the left-hand side does not
affect the leading term of the asymptotic expression. Indeed, the contribution of each term of the
form (55) grows in this case no faster than C1 T M −2 (C1 is a constant); the number of such terms is
bounded by a constant that depends on the graph (at each vertex of valency v, there are at most 2v
such terms, and since the number of vertices is finite, we obtain a finite number of terms of order
at most T M −2 ). This yields the required asymptotic formula for the number N (T ).
Remark. If a graph has a complicated structure, then the problem of writing out formulas for
the leading term of the asymptotic expression becomes much more laborious. It seems that in each
specific case one needs to enumerate all tree subgraphs in which a vertex of maximal valency has
degree smaller by one than its degree in the original graph. In addition to the contribution given by
considering this vertex as a star vertex (see Proposition 3.1), one should take into account the num-
ber of quantum packets “transferred” by this vertex to other vertices of maximal valency. Namely,
when considering conditions of the form q1 l1 + . . . + qs ls ≤ T (see the proofs of Propositions 3.1
and 3.2), where li are the travel times of composite edges and qi are positive integers, one should
take into account not only even qi (as in the case of a star graph) if the edge connects two vertices
of maximal degree. Thus, the leading term of the asymptotic expression depends on (i) the travel
time along each edge in the graph and (ii) the topological structure of the subgraph connecting the
vertices of maximal valency.
A general scheme for determining the number of packets at time T is as follows: (i) by a graph
with a given edge travel times, one constructs a certain expanding polyhedron (in the case of a star
graph, this is a simplex), and (ii) one calculates the number of integer points that lie on certain
faces of this polyhedron.
3.2. Density of distribution of packets.
Theorem 3.2 (cf. [24]). Consider the graph mentioned in Proposition 3.2 in the case of v = 3
(it consists of two vertices a and b connected by three edges; see Fig. 3). On one of the edges, consider
a segment cd with travel time τ . Then, for almost all incommensurable t1 , t2 , t3 (see [23]), the

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


260 V.L. CHERNYSHEV

c d
a b

Fig. 3. Graph and a distinguished segment cd on it.

ratio of the number of quantum packets on this segment to the number of quantum packets on the
entire graph tends to the expression

σ3 ω1 ω2 ω3 1
τ≡ τ≡ τ. (56)
σ2 ω1 ω2 + ω1 ω3 + ω2 ω3 t1 + t2 + t3

Remark. We see that quantum packets are distributed (for a given potential and given initial
conditions) uniformly with respect to the edge travel time. Of course, this does not mean that the
packets are uniformly distributed with respect to the spatial coordinate.
Proof of Theorem 3.2. The number of points on the segment cd at time instant T (denote
it by Ncd (T )) can be represented as

Ncd (T ) = N→c (T ) − N→c (T − τ ) + N→d (T ) − N→d (T − τ ); (57)

N→c (T ) stands for the number of quantum packets that passed through the point c from vertex a.
Let the travel time along the segment ac be T1 .
Let us show that
1
N→c (T ) = ω1 ω2 ω3 (T − T1 )3 + y(ω1 , ω2 , ω3 , T1 )(T − T1 )2 + o(T 2 ). (58)
12
The explicit expression of the function y(ω1 , ω2 , ω3 , T1 ) is not important for us.
Notice that N→c (T ) is equal to the number of quantum packets that came from the vertex a
along the given edge at time T − T1 . Most often, quantum packets go out from the vertex a at the
instants when packets arrive at this vertex along all the three edges (and therefore the number of
packets is not changed). Let us evaluate the number of such situations.
The quantum packets that left the vertex a, traveled along all the three edges, and returned to
the same vertex are described (see the proof of Proposition 3.1) by the inequalities

2t1 x + 2t2 y + 2t3 z ≤ T,


(2x + 1)t1 + (2y + 1)t2 + 2zt3 ≤ T,
(2x + 1)t1 + yt2 + (2z + 1)t3 ≤ T,
2xt1 + (2y + 1)t2 + (2z + 1)t3 ≤ T.

Each of these inequalities gives a contribution

1 T3 1
+ yj (t1 , t2 , t3 )T 2 + o(T 2 ) = ω1 ω2 ω3 T 3 + yj (t1 , t2 , t3 )T 2 + o(T 2 ).
3! 2t1 2t2 2t3 48

Here we took into account that the number of integer points that fall into a simplex with the sides
t1 T , t2 T , and t3 T can be expressed as t1 t2 t3 T 3 /6 + y(t1 , t2 , t3 )T 2 + o(T 2 ) for almost all t1 , t2 , t3
(see [23] and references therein).

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


TIME-DEPENDENT SCHRÖDINGER EQUATION 261

Thus, we find that the number of packets traveling from the vertex a toward the point c is
given by
4 1
ω1 ω2 ω3 T 3 + y(t1 , t2 , t3 )T 2 + o(T 2 ) = ω1 ω2 ω3 T 3 + y(t1 , t2 , t3 )T 2 + o(T 2 ).
48 12
We have proved formula (58). Substituting it into formula (57), we obtain (suppose, for defi-
niteness, that the segment cd is taken on the first edge)

1 
Ncd (T ) = ω1 ω2 ω3 (T − T1 )3 − (T − T1 − τ )3 + (T + T1 + τ − t1 )3 − (T + T1 − t1 )3
12

+ yc (ω1 , ω2 , ω3 , T1 ) (T − T1 )2 − (T − T1 − τ )2

+ yd (ω1 , ω2 , ω3 , T1 ) (T + T1 + τ − t1 )2 − (T + T1 − t1 )2 + o(T 2 ). (59)

After obvious cancellations, we arrive at


1 2
Ncd (T ) = τ T ω1 ω2 ω3 + o(T 2 ).
2
It remains to divide this expression by the total number (obtained in Proposition 3.2) of quantum
packets on the entire graph.
Remark. A condition that guarantees that the second term of the asymptotic expression for
the given set t1 , t2 , t3 has the required form is that this set is poorly approximated by rational
numbers.
Together with O.V. Sobolev, we carried out a numerical simulation for a number of finite graphs.
The experiment confirmed the correctness of the results obtained.

ACKNOWLEDGMENTS
The author is grateful to A.I. Shafarevich, A.V. Borovskikh, V.L. Pryadiev, P.B. Kurasov,
M.M. Skriganov, N.G. Moshchevitin, O.M. Kasim-Zade, Yu.V. Nesterenko, R.R. Gontsov, O.V. So-
bolev, and A.I. Barvinok for useful discussions and a number of valuable remarks. The author is
also grateful to the administration of Laboratoire de Mécanique, Modélisation et Procédés Propres,
CNRS, for providing computational facilities.
This work was supported in part by grants of the President of the Russian Federation (project
nos. MK-943.2010.1 and NSh-3224.2010.1), by the Ministry of Education and Science of the Rus-
sian Federation under the program “Development of the Scientific Potential of Higher Learning
Institutions” (project no. 2.1.1/227), and by the Russian Foundation for Basic Research (project
nos. 09-07-00327-a and 10-07-00617-a).

REFERENCES
1. A. V. Borovskikh and A. V. Kopytin, “Propagation of Waves on Networks,” in Collection of Papers of Postgrad-
uates and Students of the Faculty of Mathematics, Voronezh State University (Voronezh. Gos. Univ., Voronezh,
1999), pp. 21–25 [in Russian].
2. N. I. Gerasimenko and B. S. Pavlov, “Scattering Problems on Noncompact Graphs,” Teor. Mat. Fiz. 74 (3),
345–359 (1988) [Theor. Math. Phys. 74, 230–240 (1988)].
3. N. V. Glotov, “Differential Equations on Geometric Graphs with Singularities in the Coefficients,” Cand. Sci.
(Phys.–Math.) Dissertation (Voronezh State Univ., Voronezh, 2007).
4. N. V. Glotov and V. L. Pryadiev, “Description of the Solutions to a Wave Equation on a Finite and Bounded
Geometric Graph under Transmission Conditions of ‘Liquid’ Friction Type,” Vestn. Voronezh. Gos. Univ., Ser.
Fiz. Mat., No. 2, 185–193 (2006).

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010


262 V.L. CHERNYSHEV

5. M. G. Zavgorodnii, “Spectral Completeness of the Root Functions of a Boundary Value Problem on a Graph,”
Dokl. Akad. Nauk 335 (3), 281–283 (1994) [Russ. Acad. Sci., Dokl. Math. 49 (2), 298–302 (1994)].
6. A. V. Kopytin and V. L. Pryadiev, “On a Representation of Solutions to the Wave Equation on a Network,” in
Modern Methods in the Theory of Functions and Related Problems: Abstracts of Papers (Voronezh. Gos. Univ.,
Voronezh, 2001), p. 313.
7. V. P. Maslov, The Complex WKB Method for Nonlinear Equations (Nauka, Moscow, 1977). Engl. transl. of the
first part: V. P. Maslov, The Complex WKB Method for Nonlinear Equations. I: Linear Theory (Birkhäuser,
Basel, 1994).
8. V. P. Maslov and M. V. Fedoryuk, Quasiclassical Approximation for Equations of Quantum Mechanics (Nauka,
Moscow, 1976). Engl. transl.: V. P. Maslov and M. V. Fedoryuk, Semi-classical Approximation in Quantum
Mechanics (D. Reidel, Dordrecht, 1981).
9. Yu. V. Pokornyi, O. M. Penkin, V. L. Pryadiev, A. V. Borovskikh, K. P. Lazarev, and S. A. Shabrov, Differential
Equations on Geometric Graphs (Fizmatlit, Moscow, 2004) [in Russian].
10. Yu. V. Pokornyi, V. L. Pryadiev, and A. V. Borovskikh, “The Wave Equation on a Spatial Network,” Dokl.
Akad. Nauk 388 (1), 16–18 (2003) [Dokl. Math. 67 (1), 10–12 (2003)].
11. V. L. Pryadiev, “Description of Solutions to the Initial–Boundary-Value Problem for a Wave Equation on a One-
Dimensional Spatial Network in Terms of the Green Function of the Corresponding Boundary-Value Problem for
an Ordinary Differential Equation,” Sovrem. Mat. Prilozh. 38, 82–94 (2006) [J. Math. Sci. 147 (1), 6470–6482
(2007)].
12. V. L. Chernyshev and A. I. Shafarevich, “Semiclassical Spectrum of the Schrödinger Operator on a Geometric
Graph,” Mat. Zametki 82 (4), 606–620 (2007) [Math. Notes 82, 542–554 (2007)].
13. F. Ali Mehmeti, Nonlinear Waves in Networks (Akademie Verlag, Berlin, 1994), Math. Res. 80.
14. A. Barvinok, “Computing the Ehrhart Quasi-polynomial of a Rational Simplex,” Math. Comput. 75, 1449–1466
(2006).
15. C. Cattaneo and L. Fontana, “D’Alambert Formula on Finite One-Dimensional Networks,” J. Math. Anal. Appl.
284 (2), 403–424 (2003).
16. E. Ehrhart, “Sur les polyèdres rationnels homothétiques à n dimensions,” C. R. Acad. Sci. Paris 254, 616–618
(1962).
17. P. Exner and O. Post, “Convergence of Spectra of Graph-like Thin Manifolds,” J. Geom. Phys. 54, 77–115 (2005).
18. G. H. Hardy and J. E. Littlewood, “Some Problems of Diophantine Approximation: The Lattice-Points of a
Right-Angled Triangle,” Proc. London Math. Soc., Ser. 2, 20, 15–36 (1921).
19. Integer Points in Polyhedra—Geometry, Number Theory, Algebra, Optimization (Am. Math. Soc., Providence,
RI, 2005), Contemp. Math. 374.
20. P. Kuchment, “Quantum Graphs: An Introduction and a Brief Survey,” in Analysis on Graphs and Its Applica-
tions (Am. Math. Soc., Providence, RI, 2008), Proc. Symp. Pure. Math. 77, pp. 291–312.
21. P. Kurasov and M. Nowaczyk, “Inverse Spectral Problem for Quantum Graphs,” J. Phys. A 38, 4901–4915
(2005).
22. G. Lumer, “Connecting of Local Operators and Evolution Equations on Networks,” in Potential Theory: Proc.
Colloq. Copenhagen, 1979 (Springer, Berlin, 1980), Lect. Notes Math. 787, pp. 219–234.
23. M. M. Skriganov, “Ergodic Theory on SL(n), Diophantine Approximations and Anomalies in the Lattice Point
Problem,” Invent. Math. 132 (1), 1–72 (1998).
24. V. L. Chernyshev and A. I. Shafarevich, “Semiclassical Asymptotics and Statistical Properties of Gaussian
Packets for the Nonstationary Schrödinger Equation on a Geometric Graph,” Russ. J. Math. Phys. 15 (1), 25–34
(2008).

Translated by I. Nikitin

PROCEEDINGS OF THE STEKLOV INSTITUTE OF MATHEMATICS Vol. 270 2010

You might also like