Sie sind auf Seite 1von 8

Supplemental Information: A Tractable Numerical Model

for Exploring Nonadiabatic Quantum Dynamics


Evan Camrud and Daniel B. Turner*

Department of Chemistry, New York University, 100 Washington Square East, New York
5 NY 10003, USA

Forming the Hilbert Space


Conical intersections are naturally described in a Hilbert space composed of two

or more electronic levels and their multidimensional vibrational subspaces. In the

10 context of the current model, we reduce this very large Hilbert space to one formed by

the combination of a two-level electronic system (2LS) and a two-coordinate harmonic

oscillator (HO). Because conflicting conventions exist,S1 we explicitly write the tensor

sumS2 (⨁) used to produce the diabatic Hamiltonian in its full tensor-product (⨂) form

as

15  
⨁
 =    (S1)

=   + 

⨂ 
⨂
  (S2)

 is the identity operator in the   Hilbert space. The two-level system


where 

Hamiltonian can be written as


=  || +  ||
 (S3)

where  and  are energies of the ground (|) and excited (|) states, respectively.

Without loss of generality, we simplify the model by setting  = 0. The harmonic

20 oscillator Hamiltonian is
( (
1 1
  =  ℏ
 "
#$! + ' |$! $! | +  ℏ "
#$
+ ' |$
$
|,
(S4)
!
2

2
)* +, )- +,

where $ are the vibrational quantum numbers and "


 is the natural frequency of the

oscillator. For completeness we also write the identity operators,

(S5a)

Journal of Chemical Education 11/8/16 Page 1 of 8



= || + ||


  = ∑(
 )* +,|$! $! | + ∑)- +,|$
$
|.
(
(S5b)
 , as |1⨂|$  = |1; $ , where
We denote eigenstates of the composite Hamiltonian, 

25 1 ∈ {, }. Thus the composite Hamiltonian is



(S6)
( (
1 1
=  |; 6! ℏ "
#6! + ' ; 6! | +  |; 6
ℏ "
#6
+ ' ; 6
|
!
2

2
7* +, 7- +,

( (
1 1
+  |; $!  8 + ℏ "
#$! + '9 ; $! | +  |; $
ℏ "
#$
+ ' ; $
|.
!
2

2
)* +, )- +,

Two more steps are needed to form the complete diabatic representation of the

 ;<, = = > ?@A*BC/ℏ , where Ê! is


system. In the first step we use the displacement operator, :

the momentum operator of the first coordinate, to shift the upper oscillator an amount

<, along the first position coordinate <G! . We apply this displacement to only the excited

30 state by performing


 ! HI ;<, =
=:  :
H ;<, =, (S7)

H ;<, = = ||⨂:
where :  ;<, =. This is an adaptation of the displaced harmonic oscillator

used in models of spectroscopic signals.S3 The second step is to compute the matrix

elements describing the linear non-Condon coupling in the composite basis. The

coupling operator can be written as

 ,
L,
JG K<G L = ;|M| + |M|=⨂KN<G ,
+ O
(S8)

35 where the matrix elements of <G ,


are given by

MP|<G ,
|Q = K√Q + 1S,T! + √QS,?! L,
!

(S9)

where |P and |Q are vibrational eigenfunctions. The total diabatic Hamiltonian can be

expressed as

(S10)

Journal of Chemical Education 11/8/16 Page 2 of 8



 U 
= !
+ JG K<G L = ∑(
7* +,|; 6! ℏ
"
! V6! + W ; 6! | + ∑(
!
7- +,|; 6
ℏ
"

V6
+

W ; 6
| + ∑(
)* +,|; $!  # + ℏ V$! + W' ; $! | + ∑(
)- +,|; $
ℏ V$
+ W ; $
| +
! " ! " !
40

!

 ,
L.
;|M| + |M|=⨂KN<G ,
+ O
(S11)
We can prepare the matrix representation of the entire Hamiltonian, truncated

to a large number (X) of harmonic-oscillator eigenstates, and diagonalize to the

adiabatic basis. However, it would be challenging to distinguish the adiabatic ground

45 states from the adiabatic excited states, and the results would be numerical

approximations of both the adiabatic potentials and their vibrational eigenfunctions.

Instead we execute a similar route to the adiabatic basis by reducing the problem from

constructing and diagonalizing a 2X × 2X matrix with two coordinate dimensions to

diagonalizing a 2 × 2 matrix with a single coordinate dimension. The advantages will be

50 1. analytic expressions for the adiabatic potentials,

2. readily computable nonadiabatic coupling

3. ease of distinguishing vibrational eigenfunctions of the adiabatic ground state

from those of the adiabatic excited state.

The complication is that we need to approximate the vibrational eigenfunctions

55 numerically, and for this task we use DVR.

Analytic expressions for adiabatic states and electronic component of nonadiabatic coupling
The nonadiabatic coupling term contains both an integral over the vibrational

eigenfunctions as well as an electronic component, E[ ;<=|∇|G[ ;<=. Here we provide the

60 analytic solution to this matrix element. The electronic eigenstates are in fact vectors

that are 2 × 1 in size, which is due to the size of the electronic Hilbert space. Therefore,

to compute the matrix element, we first differentiate one vector with respect to < and

then take its dot product with the other vector. The two states are

Journal of Chemical Education 11/8/16 Page 3 of 8


(S12)

65 and

(S13)

These adiabatic electronic eigenfunctions have a complicated dependence on the

1
nuclear coordinate <. In contrast, the diabatic electronic eigenfunctions are simply ^ _
0
0
and ^ _ without any dependence on the nuclear coordinate.
1

70 After differentiation and the dot-product operation, the matrix element reduces

to

!` ;
Ha bTBC ;
cTbBC =d- =
E[ ;<=|∇|G[ ;<= = - -
* *
;cTbB=efg!`Th h qg!`Th h q (S14)
;ijkl=KmnopL ;ijkl=KmjopL

where

r = 16O
+ 4
+ 32ON< + 16N
<
+ 4 <, ;<, v 2<=

+ 4<
<,
w
v 4<<,x w
+ <,w w (S15)

and

75 y = 2 v 2<<, + <,


. (S16)

Discrete Variable Representation (DVR)


In its most simple sense, discrete variable representation is a basis

transformation, meant to utilize a known set of basis functions to approximate an

80 unknown set of basis functions. The approximate basis functions are built from a linear

combination of the known basis functions. The initial step then, is to select a set of

well-known basis functions (eigenfunctions) that will be similar in nature to the final

basis functions (eigenfunctions). For example, if one is seeking solutions to a potential

Journal of Chemical Education 11/8/16 Page 4 of 8


(2)
energy curve with sharp corners and infinite walls, a particle-in-a-box set of

85 eigenfunctions should be used. In other situations, it may be more applicable to use

other sets of basis functions (such as the solutions to the harmonic oscillator,

dampened harmonic oscillator, particle-on-a-ring, rigid rotor, and others).

Knowing the set of basis functions allows for the construction of a matrix-

representation of operators. Most forms of DVR use the position and kinetic-energy

90 operators. Because the potential-energy function is evaluated on positions—meaning we

write J;z=, and z is a position coordinate—it is helpful to be able to evaluate the

potential-energy function on a position operator. When first constructed, the position

matrix for an arbitrary basis will not be diagonal, and evaluating functions on non-

diagonal matrices is not well-defined. Therefore, one must first diagonalize the position

95 operator matrix and retain the transformation matrix used to do this. Once the diagonal

position matrix is found, the potential-energy function may be evaluated at each value

along the diagonal. These are called ‘DVR grid points’. Performing the inverse

transformation on the resulting matrix (this is why it is important to retain the

transformation matrices used when diagonalizing the position matrix) brings the

100 evaluated points back into the original basis, but they are nonetheless changed to

reflect the desired potential function. This matrix is treated as the potential-energy

matrix of the Hamiltonian, while the kinetic-energy matrix was already calculated from

the basis functions. Because of this, one may construct the final Hamiltonian matrix by

adding the kinetic and potential energies.

105 Diagonalization of this final Hamiltonian produces a matrix of eigenvectors (the

transformation matrix), and a matrix of eigenvalues (the diagonal matrix). Each

eigenvector is the collection of coefficients of the original basis functions. Essentially,

each element of the eigenvector matrix describes how much ‘weight’ each original basis

function is given. This collection of coefficients can be used to create a linear

Journal of Chemical Education 11/8/16 Page 5 of 8


110 combination of the basis functions that approximates the solutions to the potential-

energy curve given. The eigenvalues then can be used to propagate a wavepacket, or for

other purposes. In summary, the steps of DVR are as follows:

1. Choose a known, exactly solvable basis that mimics the character of the

desired potential-energy function.

115 2. Construct the matrix representation of the position (<G{ ) and kinetic-energy

( |G{
) operators in a truncated basis of } orthogonal eigenstates. The size of
!

this matrix depends on how closely the DVR solutions need to approximate

the true solutions.

3. Diagonalize <G{ to obtain the diagonal matrix <G~€ and its unitary

120 transformation matrix G.

4. Compute the potential using JK<G{ L = G JK<G~€ LG I . This produces the

potential-energy matrix in the truncated basis of } orthogonal eigenstates.

5. Construct the Hamiltonian by adding kinetic and potential energies,

{ =

!
|G
+ JK<G{ L.

7 {

125 6. Diagonalize the Hamiltonian to produce eigenvectors and eigenvalues. The

eigenvectors allow for the creation of approximate solutions to the potential.

This is done through a linear combination of the basis functions selected at

the beginning, with coefficients equal to their corresponding element in the

eigenvectors.

130 We present below the first few matrix elements of key operators for the model

system used in this work and providing in the example script,

example_avoided_crossing_script.m, which is based on the harmonic oscillator model.

The matrix elements of the position and kinetic-energy operators, <G{ and |G{
, begin
!

with

Journal of Chemical Education 11/8/16 Page 6 of 8


135

0 0.7071 0 0 0
„0.7071 0 1 0 0 ˆ
ƒ ‡
ƒ 0 1 0 1.2247 0 ‡
ƒ 0 0 1.2247 0 1.4142‡
‚ 0 0 0 1.4142 0 †

and

0.5 0 v0.7071 0 0
„ 0 1.5 0 v1.2247 0 ˆ
ƒ ‡
ƒ v0.7071 0 2.5 0 v1.7321‡,
ƒ 0 v1.2247 0 3.5 0 ‡
‚ 0 0 v1.7321 0 4.5 †

respectively. After diagonalization, separate application of the potentials, and reverse

basis transformation, the position matrices of the lower and upper adiabatic potentials

140 are

0.2468 0.0023 0.3530 v0.0001 v0.0000


„ 0.0023 0.7459 0.0030 0.6113 v0.0002ˆ
ƒ ‡
ƒ 0.3530 0.0030 1.2450 0.0035 0.8644 ‡
ƒv0.0001 0.6113 0.0035 1.7441 0.0037 ‡
‚v0.0000 v0.0002 0.8644 0.0037 2.2431 †

and

50.2532 v4.7163 0.3542 0.0001 0.0000


„v4.7163 50.7541 v6.6697 0.6134 0.0002 ˆ
ƒ ‡
ƒ 0.3542 v6.6697 51.2550 v8.1684 0.8676 ‡,
ƒ 0.0001 0.6134 v8.1684 51.7559 v9.4318‡
‚ 0.0000 0.0002 0.8676 v9.4318 52.2569 †

respectively. At this point the script maintains two separate Hamiltonian. The

Hamiltonian for the lower adiabatic potential is

0.4968 0.0023 v0.0006 v0.0001 v0.0000


„ 0.0023 1.4959 0.0030 v0.0011 v0.0002ˆ
ƒ ‡
ƒv0.0006 0.0030 2.4950 0.0035 v0.0016‡
ƒv0.0001 v0.0011 0.0035 3.4941 0.0037 ‡
‚v0.0000 v0.0002 v0.0016 0.0037 4.4931 †

145 and for the upper adiabatic potential is

50.5032 v4.7163 0.0006 0.0001 0.0000


„v4.7163 51.5041 v6.6697 0.0011 0.0002 ˆ
ƒ ‡
ƒ 0.0006 v6.6697 52.5050 v8.1684 0.0016 ‡.
ƒ 0.0001 0.0011 v8.1684 53.5059 v9.4318‡
‚ 0.0000 0.0002 0.0016 v9.4318 54.5069 †

Journal of Chemical Education 11/8/16 Page 7 of 8


The script diagonalizes these Hamiltonian matrices separately and then orders

vibrational eigenstates energetically. The five lowest-energy vibrational eigenfunctions of

the lower adiabat are

0.4968
„1.4959ˆ
ƒ ‡
ƒ2.4950‡.
ƒ3.4941‡
150

‚4.4931†

These eigenvalues demonstrate the harmonic nature of the potential to this level. The

potential remains mostly harmonic until nearly the 30th state, where the eigenvalue is

29.0150 rather than 29.5.

The five lowest-energy eigenvalues for the upper adiabat are

28.6584
„29.9706ˆ
ƒ ‡
ƒ31.3385‡.
ƒ32.7302‡
155

‚34.1255†

These eigenvalues demonstrate as well that, in addition to an energy offset of about 28,

the upper adiabat is highly anharmonic.

REFERENCES
160 S1. We have found significant confusion and misuse of the tensor sum, tensor
product, Krönecker sum, and Krönecker product in the literature. The definitions
used here are consistent with the definitive discussion and examples in Chapter
13 of Ref. S2.
S2. A. J. Laub, Matrix Analysis for Scientists and Engineers (Society for Industrial and
165 Applied Mathematics, 2005).
S3. S. Mukamel, Principles of Nonlinear Optical Spectroscopy (Oxford University Press,
New York, 1995).

Journal of Chemical Education 11/8/16 Page 8 of 8

Das könnte Ihnen auch gefallen