See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/326290418

Method for finding the exact effective Hamiltonian of time driven quantum
systems

Preprint · July 2018

CITATIONS READS

0 239

6 authors, including:

Juan Carlos Sandoval Santana V.G. Ibarra-Sierra

Universidad Nacional Autónoma de México Autonomous University of Baja California
38 PUBLICATIONS   176 CITATIONS    37 PUBLICATIONS   172 CITATIONS   

SEE PROFILE SEE PROFILE

J. L. Cardoso Alejandro Kunold

Metropolitan Autonomous University Metropolitan Autonomous University
32 PUBLICATIONS   157 CITATIONS    82 PUBLICATIONS   551 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

superlattices and heterostructures View project

Hurricane energy ranking View project

All content following this page was uploaded by Pedro Roman on 20 July 2018.

The user has requested enhancement of the downloaded file.

 Method for finding the exact effective Hamiltonian of time driven quantum systems

J. C. Sandoval-Santana,1 V. G. Ibarra-Sierra,1 J.L. Cardoso,2 A. Kunold,2 P. Roman-Taboada,3 and G. G. Naumis3

1
Departamento de Fı́sica, Universidad Autónoma Metropolitana Iztapalapa,
Av. San Rafael Atlixco 186, Col. Vicentina, 09340 Ciudad de México, México
2
Departamento de Ciencias Básicas, Universidad Autónoma Metropolitana Azcapotzalco,
Av. San Pablo 180, Col. Reynosa Tamaulipas, Ciudad de México, México
3
Departamento de Sistemas Complejos, Instituto de Fı́sica, Universidad Nacional
Autónoma de México, Apartado Postal 20-364 01000 Ciudad de México, México
Time-driven quantum systems are important in many different fields of physics like cold atoms,
solid state, optics, etc. Many of their properties are encoded in the time evolution operator which is
calculated by using a time-ordered product of actions. The solution to this problem is equivalent to
find an effective Hamiltonian. This task is usually very complex and either requires approximations,
or in very particular and rare cases, a system-dependent method can be found. Here we provide a
arXiv:1807.02550v1 [quant-ph] 6 Jul 2018

general scheme that allows to find such effective Hamiltonian. The method is based in using the
structure of the associated Lie group and a decomposition of the evolution on each group generator.
The time evolution is thus always transformed in a system of ordinary non-linear differential equa-
tions for a set of coefficients. In many cases this system can be solved by symbolic computational
algorithms. As an example, an exact solution to three well known problems is provided. For two of
them, the modulated optical lattice and Kapitza pendulum, the exact solutions, which were already
known, are reproduced. For the other example, the Paul trap, no exact solutions were known. Here
we find such exact solution, and as expected, contain the approximate solutions found by other
authors.

During the last years there has been an ever increas- posed here. As examples we can cite: the Modulated
ing interest in studying time-driven quantum systems [1] optical lattice [10, 11], Fastly driven tight-binding chains
(TDQS). Among the reasons for this spark of interest, [12, 13], Paul trap [14], Quantum wires [15], Graphene
one can mention the possibility of tailoring time driven [16], Hubbard Hamiltonian [17–19]. Furthermore, Fock
potentials using cold-atoms [2] or optically irradiated 2D space operators have the same algebra than single par-
materials [3, 4], as well as for quantum entanglement ticle Hamiltonians [1]. Therefore, if the single parti-
problems [5]. Furthermore, it has been found that new cle Hamiltonian forms a Lie algebra so does the second
and interesting topological properties arise for periodic quantization version. Therefore, the second quantization
driven systems [6]. As a matter of fact, these proper- counterpart of any single particle Hamiltonian can be ad-
ties can also be found in 2D materials, as is the case of dressed in the same way. The method can also be used
graphene [7, 8]. Also, quantum-quenching has become a to find a gauge transformation so that the Hamiltonian
mainstream subject of research [9]. In almost all of these is time-independent [1, 20].
kind of systems [1], the Hamiltonian H(t) = H0 + V (t) A Hamiltonian is said to have a dynamical algebra if
is written as a time-independent Hamiltonian (H0 ) plus it can be expressed as the superposition of the elements
a time-dependent potential (V (t)). Among the most im- of a finite Lie algebra Ln = {h1 , h2 , . . . , hn } as
portant cases, is the one of a periodic V (t). Here we will
consider such case, with V (t) having a period T . H = a> h, (2)
The TDQS properties are thus calculatedRt by using the where h = (h1 , h2 , . . . , hn ) and the coefficients a> =
time evolution operator U (t) = T e−i 0 dtH(t)/~ , where (a1 , a2 , . . . , an ) are in general time-dependent. In order
T is the time ordering operator. In the case of periodic for Ln to be a Lie algebra, any pair of its elements must
potentials, using Floquet theory, one can show that the meet the following commutator relation
solution is equivalent to find an effective Hamiltonian He n
such that [1], X
[hi , hj ] = i~ ci,j,k hk , (3)
U (T ) = e−iHe T /~ . (1) k=1

where the structure constants ci,j,k carry all the infor-

This effective Hamiltonian encodes all the dynamical in- mation regarding Ln . Part of this information concerns
formation of the system, yet its calculation is not a triv- how the unitary group generated by L transforms any
ial task. In fact, many few cases allow a closed analytic hk ∈ Ln . Indeed, it can be shown that these transfor-
solution [1]. The reason of such difficulty is that usu- mations depend entirely on the structure constants. The
ally, H0 and V (t) do not commute. Here we present a elements of the unitary group Uk = exp (iαk hk /~) trans-
general method based on the use of Lie algebras that form h according to
allows to compute He . A great variety of physically rele-
vant Hamiltonians may be addressed by the method pro- Uk hUk† = Mk h. (4)
 2

†
The matrices Mk can be calculated by taking the deriva- In order for UA to be the evolution operator, the con-
tive of the left-hand side of (4) with respect to the pa- dition u = 0 must be fulfilled [21]. This condition
rameter translates into a system of ordinary differential equations
(ODE) for the α(t) parameters that one could in princi-
i
∂αk Uk hUk† = Uk [hk , h] Uk† = −Qk Uk hUk† , (5) ple attempt to solve. However, specially for algebras with
~ large dimension, these equations might be very complex.
where the matrix elements of Qk are related to the struc- Therefore, instead, we solve the simpler system of differ-
ture constants by (Qk )i,j = ci,j,k . By using the condition ential equations
Uk hUk† = h for αk = 0, the formal solution to the differ-
ential equation (5) is given by E = ν −1 u = ν −1 Mn> . . . M2> M1> a − α̇ = 0. (14)

To insure that UA [α(0)] = 1, the initial condition α(0) =

Uk hUk† = exp (−Qk αk ) h, (6)
0 must be applied. Determining α(t) allows us to fully
and therefore, the explicit form of the transformation ma- express the evolution operator in the form (9). In order to
trices in Eq. (4) is given by find the effective Hamiltonian, the so obtained evolution
operator must be put in the form of UB . Finding the re-
Mk = exp (−Qk αk ) . (7) lation between α(t) and β(t) is then essential to working
out the effective Hamiltonian. To obtain such a relation
The time evolution operator pt = i~∂/∂t is trans- we start by assuming that both forms of the evolution
formed as operator, (9) and (10), coincide. This equality should be
h i preserved if we introduce a dependence in an auxiliary
Uk pt Uk† = pt + Uk pt , Uk† = pt + α> Ik h, (8) parameter λ by making UA [α(λ, t)] = UB (λβ(t)). It is
important to stress that at this point α(λ, t) is both a
where (Ik )i,j = δi,j δk,j . The general form of the evolu- function of the parameter λ and time. Conversely, β(t)
tion operator U (t) for a Hamiltonian with a dynamical is strictly a function of time. When λ = 0, α(0, t) = 0
algebra, can be expressed in terms of either of the follow- since UB (0) = UA [α(0, t)] = 1. Furthermore, for λ = 1
ing two forms we recover the original parameters α(1, t) = α(t). Tak-
1 1 ing the derivative with respect to λ of both sides of the
Y Y
UA (α) = Uk = exp (iαk hk /~) , (9) previous equation we get
k=n k=n
∂λ UA [α(λ, t)] = [∂λ α> (λ, t)]ν > h
n
!  
i X i >
UB (β) = exp βk hk = exp β h , (10) = β > (t)h = ∂λ UB (λβ(t)), (15)
~ ~
k=1

where U (t) = UA †
= UB †
, α> = (α1 , α2 , . . . , αn ) and where ν ≡ ν[α(λ, t)]. Factorizing h, transposing and
>
β = (β1 , β2 , . . . , βn ) are in general time-dependent pa- inverting ν, Eq. (15) can be recast in the form of the
rameters yet to be determined. We readily notice that ODE system of differential equations for α(λ, t)
the evolution operator in (10) has the form of (1) and
∂λ α(λ, t) = ν −1 [α(λ, t)]β(t). (16)
therefore it follows that
The key element to deduce the relation between α(t) and
β > (T )h/T = He . (11)
β(t) is solving this ODE system. Its solution renders
Even though in principle it would seem that a direct path α(λ, t) in the form of a function of λ and β(t)
to obtain He is to workout the β(t) coefficients, the differ-
α(λ, t) = α(λ, β(t)). (17)
ential equations that arise from the evolution operator in
(10) are extremely complicated. Fortunately, the differ- The inverse of (17) evaluated in λ = 1 yields the desired
ential equations ensued from UA are simpler and render relation of β(t) as a function of α(1, t)
the α(t) parameters instead. This, nevertheless, requires
that a relation between the α(t) and β(t) parameters be β(t) = β[α(1, t)] = β[α(t)]. (18)
established.
We thus start by determining the α(t) parameters. Af- Nonetheless, the analytical solution of the ODE system
ter successively applying the n transformations in (9) to (16) or the inverse relation (18) might be challenging to
the Floquet operator H − pt [21] and using (4) and (8), work out. To overcome this difficulty we observe that
† † † †
UA (H − pt )UA = UA (a> h − pt )UA = u> h − pt , where β > (t)h = UB β > (t)hUB = β > (t)UA hUA = β > (t)Ma h,
where
u> = a> M1 M2 . . . Mn − α̇> (t)ν, (12)
>
ν = I1 M2 . . . Mn + I2 M3 . . . Mn + · · · + In . (13) M a = M 1 M 2 . . . Mn . (19)
 3

By factorizing h and transposing we find that where C(a, q, ωt/2) is the even Mathieu function with
a = 4ω12 /ω 2 and q = −2ω02 /ω 2 . In order to obtain the
Ma> β(t) = β(t). (20)
β(t) we derive the ODE system for λ from (16)
This means that β(t) is any eigenvector of Ma> with
eigenvalue equal to 1, therefore, in general ∂λ α1 (λ, t) = β1 (t)e−4α2 (λ,t) , (26)
m
X ∂λ α2 (λ, t) = β2 (t) − 2β1 (t)α3 (λ, t), (27)
β(t) = γk (t)ρk (t), (21) ∂λ α3 (λ, t) = 4β1 (t)α32 (λ, t) + β3 (t)
k=1
−4β2 (t)α3 (λ, t). (28)
where γk (t) are coefficients to be determined and ρk (t)
are the eigenvectors of Ma> whose eigenvalues are 1. This To avoid solving the whole system of differential equa-
equation directly provides a relation between the com- tions we may use the only eigenvalue one eigenvector of
ponents of β(t) and the α(t) and reduces the search of Ma> , given in the SM. Therefore
parameters to γ1 (t), . . . , γm (t) where m < n.
α1 (t) 4α1 (t)α3 (t) − e−4α2 (t) + 1
 
Summarizing, the method to determine He works as β(t) = γ1 (t) , ,1 ,
follows. 1) Calculate the time-dependent α(t) parame- α3 (t) 4α3 (t)
ters by using Eq. (14) with the initial condition α(0) = 0. (29)
2) Connect α(t) and β(t) by means of the solution of the where the explicit form of γ1 (t) is given in the SM. Sub-
ODE system (16) in the form (18) and, if necessary, use stituting the three components of β we finally obtain the
the eigenvalue one eigenvectors of Ma> in Eq. (21) to sim- effective Hamiltonian
plify the inverse relation (18). 3) Finally, He is obtained 
γ1 (T ) 2 α1 (T ) 2
from (11). He = p + x
T α3 (T )
In what follows, we apply the method to three well
4α1 (T )α3 (T ) − e−4α2 (T ) + 1

known problems: for the first one (Paul trap), only ap- + (xp + px) . (30)
proximate solutions are known and the last two of them 4α3 (T )
(modulated optical lattice and the Kapitza pendulum)
have closed solutions. Here we find exact solutions for To first order in H0 (H0  V ) the effective Hamil-
the three of them. As this method is rather systematic, tonian is given by He = p2 /2m + x2 mω04 /4ω 2 (SM)
it can be put in the form of a symbolic computational in full consistency with [1]. Even though the effective
algorithm in Mathematica [22]. The algorithms are pro- Hamiltonian in Eq. (30) is exact, it can be recast in
vided in the supplemental material (SM) [23]. a more suitable form as to allow the computation of
Example 1: Paul trap - Ion traps use time-dependent the quasi-energies. Applying the unitary transformation


electric fields in the radio frequency domain [1, 14] to U ≡ U1 (β2 (t)/2β3 (t)) = exp ix2 β2 (t)/2β3 (t)~ the effec-
confine charged ions. They are often studied through tive Hamiltonian is transformed into
the Hamiltonian of a particle of mass m in a modulated β 2 (T ) 2
 
β3 (T ) 2 1
harmonic potential He0 = U He U † = p + β1 (T ) − 2 x ,
T T β3 (T )
1 2 m 2 (31)
ω1 + ω02 cos(ωt) x2 . (22)

H = H0 + V (t) = p + where β1 (T ), β2 (T ) and β3 (T ) are readily obtained from
2m 2
The natural frequencies of the constant and modulated (29). Figures 1 (a) and (b) exhibitp the behaviour of
potentials are ω1 and ω2 , respectively, and ω is the ra- the effective energy ~Ω/~ω = β1 β3 − β22 /π and mass
dio angular frequency. It can be easily shown that the M/m = π/mωβ3 as functions of the drive’s frequency
operators that constitute (22) form a Lie algebra. The ω0 /ω. The green solid lines show the exact calcula-
commutators of h1 = x2 , h2 = p 2 and h3 = xp + px are tions and the blue ones show the results correspond-
 2 2
x , p = i~2 (xp + px) = h3 , x2 , xp + px = i~4x2 = ing to the approximation
√ H0  V , M/m = 1 and

h1 , p2 , xp + px = −i~4p2 = h2 . Hence, its structure ~Ω/~ω = ω02 / 2ω 2 . We observe that for small values of
constants are c1,2,3 = −c2,1,3 = 2, c1,3,1 = −c3,1,1 = 4 ω the exact and approximate solutions of Ω/~ω slightly
and c2,3,1 = c3,2,1 = −4. This algebra corresponds to the diverge. The exact effective mass, on the other hand,
generators of the SU(2) group [24]. is rather different from the approximated one, even for
As shown in the SM, the solution resulting from the small values of ω/ω0 .
ODE time-dependent transformation parameters is, Example 2: Modulated optical lattice- The second-
quantized tight-binding Hamiltonian of the modulated
mω 2 d optical lattice [1, 11] is given by
α1 (t) = − ln C(a, q, ωt/2), (23)
8 dt
1 H = H0 + ωκ cos (ωt) V, (32)
α2 (t) = ln[C(a, q, ωt/2)/C(a, q, 0)], (24)
2 P †
C 2 (a, q, 0) t
Z
ds where κ is a constant parameter, H0 = J j (aj+1 aj +
α3 (t) = , (25)
2m 0 C 2 (a, q, ωs/2) a†j aj+1 ) is the nearest-neighbor hopping term and V =
 4

0.25 c4,3,2 = −c3,4,2 = 2m2 ω02 . By following the method, as

0.24
detailed in the SM, the effective Hamiltonian is
0.20 0.22

0.20

He = β1 (T ) + β2 (T )x + β3 (T )p
Ω /ω 0.15 0.18

0.16 
+ β4 (T ) p2 + (mω0 )2 x2 /T, (34)

0.48 0.50 0.52 0.54 0.56
0.10
where β1 (T ), β2 (T ), β3 (T ) and β4 (T ) are explicitly
0.05
given in the SM. This Hamiltonian can be rewrit-
ten in a more familiar form by eliminating the
0.00
0.0 0.1 0.2 0.3 0.4 0.5
terms proportional to x and p via the unitary trans-
2
ω0 / ω
formation U ≡ U2 (β3 /2β4 )U3 (−β2 /2(mω
0 ) β4 ) =
2
1.2 exp(i~xβ3 /2β4 ) exp −i~pβ2 /2(mω0 ) β4 . The trans-
formed effective Hamiltonian takes the form
1.0
p2 1 F2
He0 = U He U † = + mω02 x2 + . (35)
0.8 2m 2 4m (ω 2 − ω02 )
M /m

0.6 Though this effective Hamiltonian has not been deter-

mined explicitly before, (35) is consistent with its very
0.4
well known quasienergies [20].
0.2 In conclusion, we have presented a general method to
find the time evolution operator and the effective Hamil-
0.0
0.0 0.1 0.2 0.3 0.4 0.5
tonian for time-driven systems using an algebraic ap-
ω0 / ω
proach. Then we reproduced the solutions for known ex-
act solvable models, while we solved the Paul trap model.
Figure 1. Paul trap effective energy ~Ω (a) and effective mass This work was supported by DCB UAM-A grant num-
M (b) as function of the drive’s frequency ω0 /ω obtained from bers 2232214 and 2232215, and UNAM DGAPA PAPIIT
Eq. (31). The green solid curves show the exact results and IN102717. J.C.S.S. has a scholarship from Becas de Pos-
the blue solid curves show the approximation at first order grado UAM number 2151800745.
(H0  V ).

ja†j aj is the lattice potential. The operators a†j and a†j

P
j
are standard boson creation and annihilation operators [1] N. Goldman and J. Dalibard, Phys. Rev. X 4, 031027
at cite j. Following the procedure described above the (2014).
effective Hamiltonian is found to be [2] G. Carleo, L. Cevolani, L. Sanchez-Palencia, and
M. Holzmann, Phys. Rev. X 7, 031026 (2017).
  [3] F. J. López-Rodrı́guez and G. G. Naumis, Phys. Rev. B
He = β1 (T )h1 +β2 (T )h2 +β3 (T )h3 /T = J0 (κ)H0 , 78, 201406 (2008).
[4] G. Usaj, P. M. Perez-Piskunow, L. E. F. Foa Torres, and
where β1 (T ) = β2 (T ) = 0, and β3 (T ) = T J0 (κ). A C. A. Balseiro, Phys. Rev. B 90, 115423 (2014).
detailed calculation of these parameters can be found in [5] A. Nahum, J. Ruhman, S. Vijay, and J. Haah, Phys.
the SM. He is the same as the exact solution given in Rev. X 7, 031016 (2017).
[6] P. Roman-Taboada and G. G. Naumis, Journal of Physics
Ref. [1]. Communications 1, 055023 (2017).
Example 3: Kapitza pendulum- Here we examine the [7] T. Low, Y. Jiang, M. Katsnelson, and F. Guinea, Nano
Hamiltonian of a harmonic oscillator subject to a time- letters 12, 850 (2012).
dependent force [20] [8] G. G. Naumis, S. Barraza-Lopez, M. Oliva-Leyva, and
H. Terrones, Reports on Progress in Physics 80, 096501
p2 1 (2017).
H= + mω02 x2 + xF cos(ωt). (33) [9] E. Guardado-Sanchez, P. T. Brown, D. Mitra, T. De-
2m 2
vakul, D. A. Huse, P. Schauß, and W. S. Bakr, Phys.
In principle, the three elements in this Hamiltonian can Rev. X 8, 021069 (2018).
be identified as part of the algebra formed by the operator [10] D. H. Dunlap and V. M. Kenkre, Phys. Rev. B 34, 3625
set h1 = 1, h2 = x, h3 = p , h4 = x2 , h5 = xp + px, (1986).
[11] H. Lignier, C. Sias, D. Ciampini, Y. Singh, A. Zenesini,
h6 = p2 However, calculations are sizeabley simplified O. Morsch, and E. Arimondo, Phys. Rev. Lett. 99,
by choosing instead h1 = 1, h2 = x, h3 = p , h4 = 220403 (2007).
m2 ω02 x2 +p2 . The corresponding non-vanishing structure [12] A. Itin and A. Neishtadt, Physics Letters A 378, 822
constants are c2,3,1 = −c3,2,1 = 1, c4,2,3 = −c2,4,3 = −2, (2014).
 5

[13] T. Čadež, R. Mondaini, and P. D. Sacramento, Phys. 111, 175301 (2013).

Rev. B 96, 144301 (2017). [19] A. Eckardt, Rev. Mod. Phys. 89, 011004 (2017).
[14] Avan, P., Cohen-Tannoudji, C., Dupont-Roc, J., and [20] S. Rahav, I. Gilary, and S. Fishman, Phys. Rev. A 68,
Fabre, C., J. Phys. France 37, 993 (1976). 013820 (2003).
[15] L. Jiang, T. Kitagawa, J. Alicea, A. R. Akhmerov, [21] J. C. Sandoval-Santana, V. G. Ibarra-Sierra, J. L. Car-
D. Pekker, G. Refael, J. I. Cirac, E. Demler, M. D. Lukin, doso, and A. Kunold, Journal of Mathematical Physics
and P. Zoller, Phys. Rev. Lett. 106, 220402 (2011). 57, 042104 (2016).
[16] A. Eckardt and E. Anisimovas, New Journal of Physics [22] I. Wolfram Research, “Mathematica 11.0.1.0,” (Cham-
17, 093039 (2015). paign, IL 61820-7237, USA, 1988-2015).
[17] A. Eckardt, C. Weiss, and M. Holthaus, Phys. Rev. Lett. [23] See Supplemental Material at [URL will be inserted by
95, 260404 (2005). publisher] for a ...
[18] A. Verdeny, A. Mielke, and F. Mintert, Phys. Rev. Lett. [24] C. M. Cheng and P. C. W. Fung, Journal of Physics A:
Mathematical and General 21, 4115 (1988).

View publication stats