Molecular Physics

An International Journal at the Interface Between Chemistry and

Physics

ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/tmph20

Propagators for molecular dynamics in a magnetic

field

Laurens D. M. Peters, Erik I. Tellgren & Trygve Helgaker

To cite this article: Laurens D. M. Peters, Erik I. Tellgren & Trygve Helgaker (2024) Propagators
for molecular dynamics in a magnetic field, Molecular Physics, 122:5, e2259008, DOI:
10.1080/00268976.2023.2259008

To link to this article: https://doi.org/10.1080/00268976.2023.2259008

© 2023 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group.

Published online: 25 Sep 2023.

Submit your article to this journal

Article views: 259

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tmph20
MOLECULAR PHYSICS
2024, VOL. 122, NO. 5, e2259008 (12 pages)
https://doi.org/10.1080/00268976.2023.2259008

RESEARCH ARTICLE

Propagators for molecular dynamics in a magnetic field

Laurens D. M. Peters , Erik I. Tellgren and Trygve Helgaker
Hylleraas Centre for Quantum Molecular Sciences, Department of Chemistry, University of Oslo, Oslo, Norway

ABSTRACT ARTICLE HISTORY

Ab initio molecular dynamics in a magnetic field requires solving equations of motion with velocity- Received 14 June 2023
dependent forces – namely, the Lorentz force arising from the nuclear charges moving in a magnetic Accepted 5 September 2023
field and the Berry force arising from the shielding of these charges from the magnetic field by KEYWORDS
the surrounding electrons. In this work, we revisit two existing propagators for these equations of Molecular dynamics;
motion, the auxiliary-coordinates-and-momenta (ACM) propagator and the Tajima propagator (TAJ), magnetic field; Berry
and compare them with a new exponential (EXP) propagator based on the Magnus expansion. Addi- curvature; time propagation;
tionally, we explore limits (for example, the zero-shielding limit), the implementation of higher-order Magnus expansion
integration schemes, and series truncation to reduce computational cost by carrying out simulations
of a HeH+ model system for a wide range of field strengths. While being as efficient as the TAJ prop-
agator, the EXP propagator is the only propagator that converges to both the schemes of Spreiter
and Walter (derived for systems without shielding of the Lorentz force) and to the exact cyclotronic
motion of a charged particle. Since it also performs best in our model simulations, we conclude that
the EXP propagator is the recommended propagator for molecules in magnetic fields.

1. Introduction to determine the motion of a set of N nuclei with masses

m1 , . . . mN and positions x1 (t), . . . xN (t). The potential
Since the first simulations more than 50 years ago [1,2],
E(x) and the corresponding gradient are usually deter-
molecular dynamics has become a ubiquitous tool in
mined using force-field or (for smaller systems) ab initio
computational chemistry, allowing for the calculation of
methods, while the equations of motion are solved using
reaction rates and energies [3], the exploration of reac-
propagators such as the velocity Verlet algorithm [10,11]
tion networks [4,5], and the prediction of vibrational
or higher-order schemes [12,13].
spectra [6–9], including infrared, Raman, and circular
In a magnetic field B, a classical particle of charge
dichroism spectroscopies. The principal idea is to solve
qI experiences an additional Lorentz force:
the equations of motion
dE(x)
⎛ ⎞ mI ẍI = − − qI B × ẋI . (2)
x1 dxI
dE(x) ⎜ .. ⎟
mI ẍI = − , x = ⎝ . ⎠, (1) More than 20 years ago, Spreiter and Walter recognised
dxI
xN that the appearance of this velocity-dependent force

CONTACT Laurens D. M. Peters [email protected] Hylleraas Centre for Quantum Molecular Sciences, Department of Chemistry, University
of Oslo, P.O. Box 1033 Blindern, Oslo N-0315, Norway
© 2023 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use,
distribution, and reproduction in any medium, provided the original work is properly cited. The terms on which this article has been published allow the posting of the Accepted
Manuscript in a repository by the author(s) or with their consent.
2 L. D. M. PETERS ET AL.

requires a modification of the velocity Verlet scheme [14]. properties, such as their behaviour in the zero-field
Their algorithm, which relies on a Taylor expansion, case [see Equation (1)], in the zero-shielding case [see
has been incorporated into well-known software pack- Equation (2)], and in the cyclotron limit [only the Lorentz
ages [15,16] and has been employed in diverse applica- force in Equation (2)]. We use simulations of a HeH+
tions [17–19]. model system to corroborate our theoretical findings and
Taking into account the electrons as quantum par- test schemes to further reduce the computational cost.
ticles within the Born–Oppenheimer approximation, Having established the equations of motion, their
the energy becomes dependent on the magnetic propagation, and the weak-field limit for the step size
field [20–26], while the velocity-dependent force assumes in Sections 2.1–2.3, respectively, we derive the working
a more general form [27–29]: equations of propagators in the absence of a field as well as
the ACM, TAJ, and EXP propagators in Sections 2.4–2.7
dE(x, B) 
N
and compare them in Section 2.8. Computational details
mI ẍI = − + AIJ (x, B) ẋJ . (3) for the HeH+ simulations are given in Section 3, while the
dxI J=1
results of these simulations are presented and discussed
Here, the last term contains not only the bare Lorentz in Section 4. Conclusions and an outlook are given in
force acting on the nuclei [see Equation (2)], but also a Section 5.
contribution from the Berry curvature [30,31], reflecting
the shielding of the nuclei by the electrons [32,33] and 2. Theory
introducing a coupling between the motion of different
2.1. Equations of motion
nuclei [34,35].
While the form and the implications of Equation (3) Within the Born–Oppenheimer approximation, the
were discussed more than three decades ago [27,36,37], equations of motion of a molecule in a magnetic field are
there are only a few examples where it has been actu- given by [27–29]
ally solved for a molecular system in a magnetic field.
∂E(R, B) nucN
Ceresoli and coworkers simulated an H2 model sys- MI R̈I = − + IJ (R, B) − δIJ ZI B̃ ṘJ .
tem in 2007, integrating the equations of motion with a ∂RI
J=1
Runge–Kutta scheme [28]. In 2021, Peters and coworkers (5)
were the first to conduct accurate dynamics of H2 using
the auxiliary-coordinates-and-momenta (ACM) propa- Here, we use indices I, J, . . . for the Nnuc nuclei. RI , ZI ,
gator [29], which is derived from a general scheme for MI , and ṘI are the position, charge, mass, and velocity of
propagating nonseparable Hamiltonians by Tao [38]. A nucleus I, while the 3Nnuc -dimensional column vector R
more extensive study was conducted by Monzel and denotes the collective nuclear coordinates:
⎛ ⎞ ⎛ ⎞
coworkers [39] in 2022, studying H2 and LiH with R1 RIx
⎜ .. ⎟
the Tajima (TAJ) propagator, originating from particle R = ⎝ . ⎠ , RI = ⎝RIy ⎠ (6)
physics [40]. R
RNnuc Iz
In this work, we introduce a new propagator that we
refer to as the exponential (EXP) propagator. It is inspired We represent the uniform magnetic field B of strength B
by the fact that equations of type by the 3 × 3 antisymmetric matrix B̃ in the manner
⎛ ⎞ ⎛ ⎞
0 −Bz By Bx
ċ(t) = −iH(t) c(t), c(t0 ) = c0 , (4) ⎝
B̃ = Bz 0 −Bx , B = By ⎠ , B = |B|
⎠ ⎝
can, for a sufficiently small time interval [t0 , t], be solved −By Bx 0 Bz
exactly using the Magnus expansion [41] and approx- (7)
imated using matrix exponential(s) of −iH [42]. Such so that
equations appear, for example, in time propagations in
B̃ṘI = B × ṘI . (8)
time-dependent Kohn–Sham theory [43] and in surface-
hopping algorithms [44], where c corresponds to a vec- The first term in Equation (5) is the gradient of the
tor of state amplitudes, while H is the Hamiltonian Born–Oppenheimer electronic energy, obtained at some
matrix containing energies of and couplings between ab initio level of theory,
the states. Here, we compare the EXP propagator to the
E(R, B) = φ(R, B)|Hel (R, B)|φ(R, B) , (9)
previously published ACM and TAJ propagators, regard-
ing their theoretical foundation, their time-step require- where Hel (r, R, B) and φ(r; R, B) are the electronic
ments, their performance during actual simulations, and Hamiltonian and wave function, respectively, both of
 MOLECULAR PHYSICS 3

which depend on the electronic coordinates r [over which be integrated efficiently using different propagators.
the integration is performed in Equation (9)]. The second We denote by
term in Equation (5) is the Lorentz force arising from the
nuclear charge screened by the surrounding electrons. xa = R(t + at), (15)
This shielding is represented by the Berry curvature, π a = MṘ(t + at), (16)
whose IJ blocks,
the position and momentum, respectively, of the system
⎛ ⎞
IxJx (R, B) IxJy (R, B) IxJz (R, B) at time a relative to time t in units of the time step t.
IJ (R, B) = ⎝IyJx (R, B) IyJy (R, B) IyJz (R, B)⎠, Introducing the notation
IzJx (R, B) IzJy (R, B) IzJz (R, B)
fa = F(xa , B), (17)
(10)
wa = W(xa , B), (18)
are Cartesian matrices whose elements (here we only we may now express the equations of motion in terms of
show the xy-element), differentials with respect to the factor a
∂φ(R, B) ∂φ(R, B) ∂π a
IxJy (R, B) = −2 | , (11) π̇ a = t −1 = fa + wa π a , (19)
∂RIx ∂RJy ∂a
∂xa
are determined from derivatives of the electronic wave ẋa = t −1 = M−1 π a . (20)
∂a
function with respect to the corresponding nuclear coor-
The choice of the step length t, here appearing as a
dinates. For more details on the calculation (at the
prefactor, will be discussed in the next subsection.
Hartree–Fock level of theory) and interpretation of the
In this notation, a propagator is defined as a scheme
Berry curvature, see Refs. [30–33].
that updates the coordinates (x0 → x1 ) and momenta
With M being the 3Nnuc × 3Nnuc -dimensional matrix
(π 0 → π 1 ) for a given t. Since Equations (19) and (20)
with the nuclear masses MI on the diagonal, we can define
depend on each other, they are usually solved alternately
the kinetic momenta of the nuclei as
for a series of substeps. To ease their reading, we will
only derive working equations for a single set of those
 = MṘ. (12)
substeps,
We may now rewrite the nuclear equations of motion in a

Equation (5) in the more convenient form π0 → πa : π a = t (fα + wα π α ) dα + π 0 ,

0
(21)
˙ = F(R, B) + W(R, B),
 (13) b
x0 → xb : xb = t M−1 π β dβ + x0 , (22)
where F(R, B) is the 3Nnuc -dimensional vector of the 0
Born–Oppenheimer gradient forces and W(R, B) is a where x0 and π 0 are initial values and a and b are arbi-
3Nnuc × 3Nnuc matrix, whose IJ blocks of dimension 3 × trary step lengths in units of t. Equations for all other
3 contain the corresponding block of the Berry curvature substeps can then be derived by adjusting the initial
as well as the contribution from the bare (unscreened) values and step lengths accordingly.
Lorentz force: To evaluate the integrals in Equations (21) and (22),
we will draw inspiration from the mean-value theo-
WIJ (R, B) = MJ−1 IJ (R, B) − δIJ ZI B̃ . (14) rems. They state that, for real-valued functions ϕ(α) and
ψ(α) that are continuous in [0, a], there exists a point
The form of the equations of motion given in 0 ≤ b ≤ a such that
Equation (13) highlights that the effect of the magnetic a
field is twofold: It changes the potential energy surface ϕ(α) dα = aϕ(b), (23)
leading to different Born–Oppenheimer gradient forces 0

and it introduces an additional, velocity-dependent term. and

a b
ϕ(α)ψ(α) dt = ψ(0) ϕ(α) dα
2.2. Propagators: general considerations 0 0
a
The main objective of this theory section is to dis- + ψ(a) ϕ(α) dα. (24)
cuss how these modified equations of motion can b
4 L. D. M. PETERS ET AL.

The corresponding statements are not guaranteed to hold in some matrix norm · . Consequently, Equation (29)
for vector-valued functions; however, we will at one point holds when
rely on them as approximations. Specifically, we use the
forms t < ωa−1 . (32)
a a
This weak-field limit, as defined by Spreiter and Wal-
ϕ(α) dα ≈ ϕ(b) dα = aϕ(b), (25)
0 0 ter [14], means that the time step is small enough to
resolve the cyclotronic motion. For a neutral molecule,
and the charge–mass ratio is roughly on the order of 10−4 a.u.,
a b so that
ψ(α)ϕ(α) dα ≈ ψ(α)ϕ(0) dα
0 0 ωa ∼ 10−4 B. (33)
a
+ ψ(α)ϕ(a) dα, (26) Time steps of up to 100 a.u. (2.4 fs) therefore allow for
b
simulations in field strengths up to about 10 B0 , cover-
to approximate integrals over two vectors of continuous ing the entire range of field strengths from the (particu-
functions ϕ(α) and ψ(α). Note that both the midpoint larly interesting) intermediate regime (< 1 B0 ) up to the
rule and the trapezoidal rule can be derived from these Landau regime.
mean-value approximations by choosing b = a/2 and, in
case of latter, ψ(α) = 1:
2.4. Propagators in the absence of a field
a a
ϕ(α) dα ≈ ϕ(a/2) dα = aϕ(a/2) (27) For the field-free case, w is zero so that we can directly
0 0 apply the mean-value approximation Equation (25) to
a a/2 Equations (21) and (22):
ϕ(α) dα ≈ ϕ(0) dα
0 0 a
a a π a = t fα dα + π 0 ≈ at fb + π 0 (34)
+ ϕ(1) dα = [ϕ(0) + ϕ(1)] 0
a/2 2 b
(28) xb = t M−1 π β dβ + x0 ≈ bt M−1 π a + x0
0
(35)
2.3. Choice of step size
As shown in Algorithm 1, both equations are solved alter-
The error introduced by a given propagator depends on nately using the current forces and momenta as mean
the applied step size t. Before considering the differ- values to propagate the momenta and positions, respec-
ent propagators, we discuss briefly in this subsection the tively. The remaining task is now to come up with a series
restriction on the time step imposed by an external mag- of a’s and b’s (denoted by the vectors a and b) to approx-
netic field. As in field-free simulations, we demand that imate the mean values. The lengths of a and b (K + 1 and
the time step t is significantly smaller than the fastest K) reflect the order K of the scheme.
molecular vibration. In addition, we demand that the In the standard velocity Verlet scheme (K = 1)
matrix t wa is convergent for all a, meaning that [10,11], we set a = [0.5, 0.5] and b = [1.0], so that we
obtain:
lim [twa ]n = 0 (29)
n→∞
π VV
0.5 = 0.5t f0 + π 0 (36)
Introducing the spectral radius ρ  
x1VV = t M−1 π VV0.5 + x0 (37)
  
ρ waT wa = ωa , (30) π VV
1 = 0.5t f1 + π VV
0.5 (38)

which we loosely interpret as a cyclotron frequency, we Note that this quadrature corresponds to solving the full
have integral (a = 1) in Equation (34) with the trapezoidal
[see Equation (28)] and the full integral in Equation (35)
 
lim [twa ]n  (b = 1) with the midpoint rule [see Equation (27)]. As
n→∞ a consequence of this the velocity Verlet algorithm is
 n/2
symplectic, time-reversible, and of second-order accu-
∝ lim t n ρ waT wa = lim (tωa )n (31)
n→∞ n→∞ racy. However, it has been shown that higher-order
 MOLECULAR PHYSICS 5

Algorithm 1 General algorithm for a propagator of order and the propagation of π in terms of π and x
K, in the absence of a field. See Equations (15)–(18) for  
π b ≈ bt fa + wa π a + π 0 , (42)
definitions of the variables. a and b are predefined vectors
of length K + 1 and K, respectively. In case of velocity xa ≈ at M−1 π b + x0 , (43)
Verlet, a = [0.5, 0.5] and b = [1.0].
k = 0; a = 0; b = 0 The only remaining step is the coupling of coordinates
while k ≤ K do and momenta, which is done when all quantities x, π , x ,
a = a[k] and π are at the same time step by carrying out, for a
π a +a = atfb + π a given coupling constant s, the transformation [29]
a =a +a ⎛ ⎞ ⎛ ⎞
xc xc
if k < K then ⎜ xc ⎟ ⎜ ⎟
⎜ ⎟ s ⎜ xc ⎟
b = b[k]   ⎝M−1 π c ⎠ → sb ⎝M−1 π c ⎠ (44)
xb +b = bt M−1 π a + xb
M−1 π c M−1 π c
b =b +b
Calculate fb with the coupling matrix
end if ⎛ ⎞
k=k+1 1 + cos χ 1 − cos χ s−1 sin χ −s−1 sin χ
1 ⎜1 − cos χ 1 + cos χ −s−1 sin χ s−1 sin χ ⎟
end while ssb = ⎜ ⎟
2 ⎝ s sin χ −s sin χ 1 + cos χ 1 − cos χ ⎠
−s sin χ s sin χ 1 − cos χ 1 + cos χ
schemes [12,13], can significantly increase the stability of (45)
the dynamics.
Unfortunately, there is no straightforward expansion where
of the upper scheme towards systems with velocity-
χ = sbt (46)
dependent forces since, taking the velocity Verlet
algorithm as an example, the step of Equation (38) Setting s = 0 reduces the coupling matrix to the unity
becomes matrix, while setting it to s = π/(ct) leads to a com-
plete exchange of x and π with x and π and vice versa.
π 1 ≈ 0.5t (f1 + w1 π 1 ) + π 0.5 (39)
The pseudo code for a general ACM propagator of order
The mismatch between the required (π 1 ) and the avail- K is given in Algorithm 2.
able (π 0.5 ) momenta for the propagation leads to a
systematic error in the integration of the equations of 2.6. Tajima propagator
motion [29]. Clearly, there is a need to develop alter-
native propagators, which will be done in the following An alternative to the ACM propagator is the Tajima
subsections. (TAJ) propagator. Initially used in particle physics [40],
it was rewritten for molecular applications by Monzel
et al. [39]. Here, we present a slightly different deriva-
2.5. Auxiliary-coordinates-and-momenta tion of the working equations that starts with applying the
propagator mean-value approximations [Equations (25) and (26)] to
The auxiliary-coordinates-and-momenta (ACM) prop- Equation (21):
agator method was proposed by Tao for general non- c
separable Hamiltonians [38] and adapted by Peters π a ≈ at fb + t wα π 0 dα
0
and coworkers for molecular simulations in a magnetic a
field [29]. The idea is to circumvent the mismatch in + t wα π a dα + π 0 (47)
Equation (39), by introducing an additional pair of coor- c
dinates and momenta (x , π ) that are kept close to the Assuming that c = a/2 and that both integrals have the
original pair (x, π ) during the dynamics (x ≈ x and π ≈ same mean value wb , we obtain:
π ). If this coupling is sufficiently strong, we can write the
a a
propagation of π in terms of π and x 1 − t wb π a ≈ at fb + 1 + t wb π 0 (48)
2 2
π a ≈ at fb + wb π b + π 0 , (40) Introducing the inverted matrix
 
xb ≈ bt M−1 π a + x0 . (41) vbc = [1 − ctwb ]−1 , (49)
6 L. D. M. PETERS ET AL.

Algorithm 2 General algorithm for the auxiliary coor- Algorithm 3 General algorithm for the Tajima (TAJ)
dinates and momenta (ACM) method of order K. See method of order K. See Equations (15)–(18) for defini-
Equations (15)–(18) for definitions of the variables. a and tions of the variables. a and b are predefined vectors of
b are predefined vectors of length K + 1 and K, respec- length K + 1 and K, respectively. In the velocity Verlet
tively. In the velocity Verlet variant, a = [0.5, 0.5] and variant, a = [0.5, 0.5] and b = [1.0].
b = [1.0]. k = 0; a = 0; b = 0
k = 0; a = 0; b = 0 while k ≤ K do
while k ≤ K do a = a[k]  
a
a = a[k]   π a +a = vb
a/2
atfb + 1 + twb π a
π a +a = at fb + wb π b + π a 2
a =a +a
xa +a = at M−1 π b + xa
if k < K then
a =a +a b = b[k]
if k < K then  
xb +b = bt M−1 π a + xb
b = b[k] b =b +b
Calculate fa and wa  Calculate fb and wb
π a = (a − b )t fa + wa π a + π b end if
xa = (a − b )t M−1 π a + xb k=k+1
Apply ssb end while
Calculate fa and wa
b =b +b  
π b = (b − a )t fa + wa π a + π a where γ is chosen so that the Magnus series [41],
xb = (b − a )t M−1 π a + xa a

Calculate fb and wb yγ ,a = t dα wα
γ
end if α
1 a  
k=k+1 + (t)2 dα dβ wα , wβ + · · · , (53)
end while 2 γ γ

converges. From ya,a = 0 and

we arrive at the working equation for the TAJ propagator,    
  ∂ exp yγ ,a
a/2 a = t wa exp yγ ,a , (54)
π a ≈ vb atfb + 1 + twb π 0 , (50) ∂a
2
it is straightforward to verify that Equation (52) solves
yielding the algorithm in Algorithm 3. In the weak-field
Equation (19) and is thus an alternative to Equation (21).
limit [see Equation (29)], we can expand the matrix inver-
We can set γ to zero from now on, since our choice of
sion in a Neumann series:
step size in Equation (32) ensures the convergence of
∞
 
N−1
  the Magnus series in this case. Truncating the Magnus
vbc = [ctwb ]n = [ctwb ]n + O [tωb ]N series after the first term and applying the mean-value
n=0 n=0
approximation [Equation (25)], the matrix exponential
(51)
becomes
In standard applications, we expect this series to converge    
fast. Using the estimate in Equation (33) with B = 0.1 exp yα,a ≈ exp [a − α] twb = ua−α b , (55)
and t = 10 in atomic units, the error is on the order of where we have introduced the matrix exponential
10−8 when using N = 2. Consequently, we can avoid the
(comparably) high computational cost of matrix inver- ub = exp(twb ) (56)
sions during the simulations. Equation (52) now becomes:
a
2.7. Exponential propagator π a ≈ t ua−α
b dα fb + uab π 0 (57)
0
A general approach to solving first-order linear dif- Approximating the remaining integral via the midpoint
ferential equations is the Magnus integrator method. rule [Equation (27)], we obtain
Specifically, we may write
a/2 a/2 a/2
a     π a ≈ atub fb + uab π 0 = ub atfb + ub π 0
π a = t exp yα,a fα dα + exp yγ ,a π γ (52)
γ (58)
 MOLECULAR PHYSICS 7

Algorithm 4 General algorithm for the Exponential Table 1. Comparison of the auxiliary coordinates and momenta
(ACM), Tajima (TAJ), and exponential (EXP) propagator.
(EXP) method of order K. See Equations (15)–(18) for
definitions of the variables. a and b are predefined vec- Criterion ACM TAJ EXP
tors of length K + 1 and K, respectively. In the velocity Forces calculations per step 3K K K
Parameter-free? No Yes Yes
Verlet variant, a = [0.5, 0.5] and b = [1.0]. Correct zero-field limit? No Yes Yes
k = 0; a = 0; b = 0 Correct zero-shielding, No Yes Yes
velocity Verlet limit?
while k ≤ K do Exact cyclotron limit? No No Yes
a = a[k]
a/2
π a +a = atub fb + uab π a
a =a +a external force f and a magnetic field of strength Bz in the
if k < K then z-direction. In this particular case, w and thus ua and va
b = b[k] are constants
  ⎛ ⎞
xb +b = bt M−1 π a + xb 0 ω 0
Bz Z1
b =b +b w = ⎝−ω 0 0⎠ ω = , (60)
M1
Calculate fb and wb 0 0 0
end if and the velocity Verlet variants of the EXP and TAJ
k=k+1 propagator reduce to
end while
t  
π EXP
1 = u0.25 f1 + u0.75 f0 + u1 π 0 , (61)
2
as the exponential propagator (EXP) for the nuclear    
t t
momenta in magnetic fields; see Algorithm 4. The matrix π TAJ
1 = 0.25
v f1 + v 0.25
1+ w v f0 0.25
2 4
exponential can be expanded as    
t t
∞ + v0.25 1 + w v0.25 1 + w π 0 , (62)
 1  n 4 4
ucb = ctwb
n=0
n! respectively. Expanding ua and va up to second order
1  

N−1
1  n   ua = 1 + atw + a2 (t)2 w2 + O [t]3 (63)
= ctwb + O [tω]N /N! . (59) 2
n!  
n=0 va = 1 + atw + a2 (t)2 w2 + O [t]3 (64)
In contrast to the Neumann series, this series converges and using the Taylor expansion of the forces in
unconditionally, for all step sizes. y-direction
y
y y df0  
2.8. Comparison of the propagators f1 = f0 + t + O [t]2 , (65)
dt
We close this section by discussing a few properties of the we obtain, for both EXP and TAJ, an expression that is
introduced propagators. The results are summarised in identical to those obtained by Spreiter and Walter via
Table 1. From Algorithms 2–4, we see that the ACM prop- inversion [see Equation (16) in Ref. [14]] and via Taylor
agator is about three times more expensive than the other expansion [see Equation (39) in Ref. [14]]:
schemes, requiring 3K instead of K forces and Berry cur- t  x y
vature calculations per step. In addition, it depends on the π1x = π0x + f0 + f1x + 2ωπ0
2
definition of a parameter (s), which has an influence on 1    
y
the stability of the dynamics. Moreover, unlike the ACM + (t)2 2ωf0 − 2ω2 π0x + O [t]3 (66)
4
propagator, TAJ and EXP propagators converge to the
correct zero-field solution when setting B and therefore Their difference, however, becomes obvious when addi-
wb to zero. tionally setting the forces in Equations (61) and (62) to
As the correct zero-shielding limit, we define the prop- zero:
agators derived by Spreiter and Walter, which were con- π EXP
1 = u1 π 0 (67)
structed for the special case where the Berry curvature    
t t
is zero. While their working equations clearly differ from π TAJ
1 = v 0.25
1 + w v 0.25
1 + w π0 (68)
the ACM propagator, we can compare them to the TAJ 4 4
and EXP propagators by assuming a single particle with In this cyclotron limit, the EXP propagator collapses to the
charge Z1 and mass M1 , experiencing a time-dependent exact result for the cyclotron motion of a single, charged
8 L. D. M. PETERS ET AL.

particle energies E(dHeH ), Berry charges QIJ (dHeH ), and Berry

⎛ ⎞ charge fluctuations PIJ (dHeH ) obtained by spline fitting
cos (tω) sin (tω) 0 of Hartree–Fock (HF)/lu-aug-cc-pVTZ [46–48] results
π EXP
1 = ⎝− sin (tω) cos (tω) 0⎠ π 0 , (69) for different bond lengths (dHeH ) at this field strength.
0 0 0 The prefix ‘lu-’ indicates the use of an uncontracted
London orbital basis set, as suggested in Ref. [49].
while TAJ introduces an error at the order of O([tω]3 ). The Berry charges and Berry charge fluctuations were
For the previously mentioned estimate [see extracted from the numerical Berry curvature [30] in
Equation (33)] with B = 0.1 and t = 10, this results a least-squares fashion (see Ref. [33]). All calculations
in an error at the order of 10−12 (all given in atomic were performed with the London [50] programme
units). Thus, in practice EXP and TAJ trajectories and package. The resulting bond-length dependence of the
their computational cost will be very similar. energy and charges is shown in Figure 1. We note that
HeH+ dissociates into He with effective charge ZHe +
3. Computational details QHeHe = 0 and a proton H+ with effective charge ZH +
QHH = 1. In the bonding region, both nuclear charges
3.1. Diatomic model system are partially screened by the electrons. Note that, in
To conduct a study of the performance of the previ- our calculations, we neglect the magnetic-field depen-
ously discussed propagators, we carry out simulations of dence of the electronic energy itself E(dHeH ), of the
a diatomic model system perpendicular to the magnetic Berry charges, QIJ (dHeH ), and of the Berry fluctuations
field. It has been shown that, in this particular case, the PIJ (dHeH ). The magnetic field therefore enters the cal-
Berry curvature can be represented exactly by a set of culations only through the explicit field-dependence of
Berry charges QIJ (R, B) and Berry charge fluctuations the model Berry curvature – that is, by the depen-
PIJ (R, B) [30,33] dence on B̃ in Equation (70). This approach allows for a
more systematic study of the propagators, since only the
IJ (R, B) = −QIJ (R, B)B̃ velocity-dependent forces increase linearly with B, while
the potential-energy surface and the electronic struc-
− PIJ (R, B) B̃R̄IJ R̄IJT − R̄IJ R̄IJT B̃ , (70) ture remain unaffected. Because of its overall charge,
small mass, and significant geometry dependence of the
where R̄IJ is the normalised interatomic distance vector: screening [see Figure 1(b)], the HeH+ model system can
  be regarded as an extreme case, featuring (comparably)
R̄IJ = dIJ−1 RJ − RI dIJ = |RJ − RI | (71) large cyclotron frequencies (ω).
For a more detailed derivation and discussion of
QIJ (R, B) and PIJ (R, B), we refer the reader to
3.2. Molecular simulations
Refs. [32,33,45].
Our model system has been designed to reproduce All NVE simulations (constant number of particles, vol-
the properties of HeH+ at field strength 0.1 B0 , with ume, and energy) were conducted in the xy-plane with

Figure 1. Bond-length (dHeH ) dependent energies (a) and charges (b) derived from the Berry curvature of HeH+ at the HF/lu-aug-cc-
pVTZ level of theory perpendicular to a field of B = 0.1 B0 . We consistently use atomic units.
 MOLECULAR PHYSICS 9

Table 2. Coefficients a and b for the different integrators of order

K used in this work.
VV (K = 1)
a0 = 0.5000000000000000 b0 = 1.0000000000000000
a1 = 0.5000000000000000
OM (K = 4)
a0 = 0.1786178958448091 b0 = 0.7123418310626054
a1 = −0.0662645826698185 b1 = −0.2123418310626054
a2 = 0.7752933736500187 b2 = −0.2123418310626054
a3 = −0.0662645826698185 b3 = 0.7123418310626054
a4 = 0.1786178958448091
RK4 (K = 6)
a0 = 0.0792036964311957 b0 = 0.2095151066133620
a1 = 0.3531729060497740 b1 = −0.1438517731798180
a2 = −0.0420650803577195 b2 = 0.4343366665664560
a3 = 0.2193769557534996 b3 = 0.4343366665664560
a4 = −0.0420650803577195 b4 = −0.1438517731798180 Figure 2. Standard deviation of the total energy (εtot ) during sim-
a5 = 0.3531729060497740 b5 = 0.2095151066133620 ulations of the HeH+ model system using different magnetic field
a6 = 0.0792036964311957
strengths and propagators with K = 1 (velocity Verlet): Auxiliary
coordinates and momenta (ACM), Tajima (TAJ), and exponential
(EXP) propagator.
the magnetic field aligned along the z-axis. Each sim-
ulation began at the equilibrium geometry, with ran-
dom initial momenta. We investigated the ACM prop-
agator Algorithm 2, the TAJ propagator Algorithm 3,
4. Results and discussion
and the exponential operator Algorithm 4 at three dif-
ferent orders K: the velocity-Verlet (VV) scheme [10,11] In Figure 2, we have plotted εtot for HeH+ as a function of
with K = 1, the OM scheme of Omelyan et al. [12] with field strength B for the ACM, TAJ, and EXP propagators
K = 4, and the RK4 scheme (S6 /O4 in Ref. [13]) with with K = 1. The results for K = 4 and K = 6 are very
K = 6. The corresponding factors a and b are given in similar and therefore not included in the figure.
Table 2. In general, all three propagators become less stable
Each trajectory was simulated for 24 ps using (if not with increasing field strength – in particular the ACM
stated otherwise) an effective time step (t/K) of 0.1 fs, propagator, for which the simulation at 103 B0 becomes
storing energies, geometries, and momenta every 2.4 fs. unstable. At this high field strength (characteristic of neu-
For the ACM propagator, we used an optimised coupling tron stars rather than white dwarfs), a reoptimisation of
constant s = 0.013. Where N is not given, we used the the coupling-strength parameter s may be required. In
standard matrix-inversion and exponential algorithms agreement with a previous comparison of the ACM and
of python3.6 to calculate v and u in the TAJ and TAJ propagators [39], we note that the ACM propaga-
EXP propagators, respectively, and the truncated series tor performs less well than the TAJ and EXP propaga-
of Equations (51) and (59) otherwise. Estimating that tors. Since, in addition, the TAJ and EXP propagators
t ω = 4.5 × 10−4 B, we apply a magnetic-field range of are parameter-free and since they require only one rather
10−2 –103 in units of B0 to ensure that the weak-field than three force and Berry-curvature calculations per
condition in Equation (32) holds for all simulations. step, we conclude that the TAJ and EXP propagators are
We use the standard deviation of the total energy (εtot ) to be preferred over the ACM propagator.
as a criterion for the stability of the dynamics. In order to While the EXP and TAJ propagators are indistinguish-
estimate the cost-accuracy ratio of the methods, we also able at low field strengths, the EXP propagator becomes
study the behaviour of this error measure with respect to marginally more stable for B > 10 B0 , as the cyclotronic
the step length: centre-of-mass motion of the charged HeH+ system
begins to dominate the trajectories. However, even at
εtot ∼ p t o (72) 10 B0 , the centre-of-mass motions and the vibrational
spectra of the two propagators are still indistinguish-
The prefactor p and the order o are determined as the able and in fact identical to those obtained with the
intercept and slope of a log(εtot )-log(t)-plot, respec- ACM propagator; see Figure 3. In line with previous
tively, generated from a series of simulations with con- studies on diatomics [29,39], the peaks in the vibra-
stant B but varying t between 0.01 fs and 0.6 fs. In addi- tional spectrum of HeH+ Figure 3(b) correspond to the
tion, vibrational spectra [8,29] and the centre-of-mass cyclotronic centre-of-mass motion [∼ 30 cm−1 , also visi-
motion are used to compare trajectories of the different ble in Figure 3(a)], rotation (∼ 370 cm−1 ), and vibration
propagators. (∼ 3350 cm−1 ), and feature splitting patterns that arise
10 L. D. M. PETERS ET AL.

Figure 3. Centre-of-mass motion (a) and vibrational spectra of He (b) obtained from the simulations of the HeH+ model system at B =
10 B0 using different propagators with K = 1 (Velocity Verlet): Auxiliary coordinates and momenta (ACM), Tajima (TAJ), and exponential
(EXP) propagator.

Table 3. Orders (o) and prefactors (p) of the auxiliary coordi- from the couplings between the different modes. Select-
nates and momenta (ACM), Tajima (TAJ), and exponential (EXP) ing one set of simulations at low (B = 0.1 B0 ) and one at
propagator at two different field strenghts (B). high field strengths (B = 100 B0 ), we list the order o and
εtot ∼ to p prefactor p of εtot with respect to t in Table 3. It clearly
Propagator K B [B0 ] o p shows that all propagators with K = 1 are, just like VV
ACM 1 0.1 2.01 10−4.82 in the field-free case, of second order in both regimes.
ACM 1 100 2.00 10−3.76 Therefore, the trends of εtot in Figure 2 solely arise from
TAJ 1 0.1 2.01 10−5.10
TAJ 1 100 2.02 10−4.89 p. Note that, while o is approximately constant, p might
EXP 1 0.1 2.01 10−5.10 change for a different system.
EXP 1 100 2.04 10−5.02
EXP 4 0.1 3.96 10−7.49
Since the EXP propagator is the most stable propa-
EXP 4 100 2.05 10−5.51 gator, we restrict our attention to this propagator when
EXP 6 0.1 3.94 10−8.53 investigating dependence on the propagator order K and
EXP 6 100 2.00 10−6.25
on the truncation level N in the exponential series; see
Note: For EXP, we also list o and p of the higher-order schemes (K = 4, 6).

Figure 4. Influence of the order of the EXP propagator (K, a) and the truncation of the exponential series (N, b) on the standard deviation
of the total energy (εtot ) during simulations of the HeH+ model system using different magnetic field strengths. The reference (exact
exponential with K = 1) is shown as a dashed black line in both plots.
 MOLECULAR PHYSICS 11

Figure 4(a,b), respectively. We expect the TAJ propaga- propagator might be useful for cases where the energy
tor to show a similar behaviour, while an investigation and/or the electron shielding depend on the nuclear
of the influence of K on the ACM propagator can be momenta.
found in Ref. [29]. At low field strengths, the higher- In the study of molecules in strong magnetic fields, we
order schemes (i.e. OM with K = 4 and RK4 with K = 6) are usually interested in field strengths below 1 B0 and
improve the stability of the EXP propagator by up to two apply time steps of t = 0.1 fs to resolve the molecular
orders of magnitude; see Figure 4(a). These higher-order vibrations. In practice, these time steps are small enough
schemes therefore allow for a significantly larger effective to resolve the cyclotronic motion (weak field limit), while,
time step t/K (constant in Figure 4), reducing the com- additionally, allowing for the use of higher-order schemes
putational cost of the dynamics. Perhaps surprisingly, and truncation of the exponential series to reduce the
the OM and RK4 schemes become less stable than the computational cost of the EXP propagator.
lower-order VV scheme as the velocity-dependent forces
become dominant in the Landau regime (beyond 10 B0 ). Acknowledgments
Both observations are in line with the behaviour of o and
We thank Tanner Culpitt for helpful discussions.
p listed in Table 3. At 0.1 B0 , both OM and RK4 are of
fourth order and show a significantly smaller p, while
at 100 B0 only p is smaller when comparing to the VV Disclosure statement
scheme with K = 1. Since the latter is not enough to com- No potential conflict of interest was reported by the author(s).
pensate the increasing number of calculations per time
step, the higher-order schemes become more expensive. Funding
Both Figure 4(a) and Table 3 prove that the coefficients a
This work was supported by the Research Council of Nor-
and b (see Algorithm 4) need to be adjusted at high field way [Norges Forskningsråd] through ‘Magnetic Chemistry’
strengths to obtain o > 2 in the Landau regime. [grant number 287950] and CoE Hylleraas Centre for Quan-
Simulations with a truncated exponential series at tum Molecular Sciences [grant number 262695]. The work also
N = 2 in Equation (59) reproduce the results obtained received support from the UNINETT Sigma2, the National
with the exact matrix exponential for field strengths up Infrastructure for High Performance Computing and Data
Storage, through a grant of computer time [grant number
10 B0 ; see Figure 4. Since calculating a matrix exponential
NN4654K].
is (comparably) expensive, replacing it with a series will
reduce the computation time. For N = 2 at higher field ORCID
strengths, the truncation error exceeds O([ωt]2 ) =
Laurens D. M. Peters http://orcid.org/0000-0001-6572-8738
10−5 , giving unstable dynamics. This indicates that N
Erik I. Tellgren http://orcid.org/0000-0002-0019-4330
must be chosen with care.
Trygve Helgaker http://orcid.org/0000-0002-5032-8392

5. Conclusion and outlook References

In this work, we have studied three propagators for [1] J.B. Gibson, A.N. Goland, M. Milgram and G.H. Vine-
molecular dynamics in a magnetic field – namely, the yard, Phys. Rev. 120, 1229–1253 (1960). doi:10.1103/
auxiliary-coordinates-and-momenta (ACM) propagator, PhysRev.120.1229
[2] A. Rahman, Phys. Rev. A 136, 405–411 (1964). doi:10.
the Tajima (TAJ) propagator, and the new exponen-
1103/PhysRev.136.A405
tial (EXP) propagator, testing their performance using [3] C. Chipot and A. Pohorille, Free Energy Calculations
simulations of a HeH+ model system at different field (Springer-Verlag, Berlin, Heidelberg, 2007).
strengths (10−2 – 103 B0 ). [4] L.P. Wang, A. Titov, R. McGibbon, F. Liu, V.S. Pande
The EXP propagator, which was derived from a trun- and T.J. Martínez, Nat. Chem. 6, 1044–1048 (2014).
cated Magnus expansion, correctly reduces to the stan- doi:10.1038/nchem.2099
[5] S. Grimme, J. Chem. Theory Comput. 15, 2847–2862
dard velocity Verlet propagator in the zero-field limit, the (2019). doi:10.1021/acs.jctc.9b00143
propagators of Spreiter and Walter [14], neglecting elec- [6] R.P. Futrelle and D.J. McGinty, Chem. Phys. Lett. 12,
tron shielding, and the cyclotronic motion of a charged 285–287 (1971). doi:10.1016/0009-2614(71)85065-0
particle. Since it also performed best in our model [7] M.P. Gaigeot, M. Martinez and R. Vuilleumier, Mol. Phys.
simulations and showed the best stability, especially at 105, 2857–2878 (2007). doi:10.1080/00268970701724974
[8] M. Thomas, M. Brehm, R. Fligg, P. Vöhringer and
higher field strengths, we recommend the EXP propa-
B. Kirchner, Phys. Chem. Chem. Phys. 15, 6608–6622
gator for simulations of molecules in a magnetic field. (2013). doi:10.1039/c3cp44302g
However, the TAJ propagator delivers identical results [9] M. Thomas and B. Kirchner, J. Phys. Chem. Lett. 7,
at field strengths ≤ 1 B0 and the more expensive ACM 509–513 (2016). doi:10.1021/acs.jpclett.5b02752
12 L. D. M. PETERS ET AL.

[10] L. Verlet, Phys. Rev. 159, 98–103 (1967). doi:10.1103/ [30] T. Culpitt, L.D.M. Peters, E.I. Tellgren and T. Helgaker,
PhysRev.159.98 J. Chem. Phys. 155, 024104 (2021). doi:10.1063/5.0055
[11] W.C. Swope, H.C. Andersen, P.H. Berens and K.R. Wil- 388
son, J. Chem. Phys. 76, 637–649 (1982). doi:10.1063/1. [31] T. Culpitt, L.D.M. Peters, E.I. Tellgren and T. Helgaker,
442716 J. Chem. Phys. 156, 044121 (2022). doi:10.1063/5.0079
[12] I.P. Omelyan, I.M. Mryglod and R. Folk, Comput. Phys. 304
Commun. 151, 272–314 (2003). doi:10.1016/S0010-4655 [32] L.D.M. Peters, T. Culpitt, E.I. Tellgren and T. Helgaker,
(02)00754-3 J. Chem. Phys. 157, 134108 (2022). doi:10.1063/5.0112
[13] S. Blanes and P.C. Moan, J. Comput. Appl. Math. 142, 943
313–330 (2002). doi:10.1016/S0377-0427(01)00492-7 [33] L.D.M. Peters, T. Culpitt, E.I. Tellgren and T. Hel-
[14] Q. Spreiter and M. Walter, J. Comput. Phys. 152, 102–119 gaker, J. Chem. Theory Comput. 19, 1231–1242 (2023).
(1999). doi:10.1006/jcph.1999.6237 doi:10.1021/acs.jctc.2c01138
[15] E. della Valle, P. Marracino, S. Setti, R. Cadossi, M. Liberti [34] T. Culpitt, L.D.M. Peters, E.I. Tellgren and T. Helgaker,
and F. Apollonio, in XXXIInd Gen. Assem. Sci. Symp. Int. J. Chem. Phys. 158, 114115 (2023). doi:10.1063/5.0139
Union Radio Sci. (URSI GASS), pp. 1–4. 675
[16] K. Khajeh, H. Aminfar, Y. Masuda and M. Mohammad- [35] E.I. Tellgren, T. Culpitt, L.D.M. Peters and T. Helgaker,
pourfard, J. Mol. Model. 26, 106 (2020). doi:10.1007/ J. Chem. Phys. 158, 124124 (2023). doi:10.1063/5.0139684
s00894-020-4349-0 [36] P. Schmelcher and L.S. Cederbaum, Phys. Rev. A 40,
[17] M.S. Al-Haik and M.Y. Hussaini, Mol. Simul. 32, 601–608 3515–3523 (1989). doi:10.1103/PhysRevA.40.3515
(2006). doi:10.1080/08927020600887781 [37] P. Schmelcher and L.S. Cederbaum, Int. J. Quantum
[18] K.T. Chang and C.I. Weng, J. Appl. Phys. 100, 043917 Chem. 64, 501–511 (1997). doi:10.1002/(SICI)1097-461X
(2006). doi:10.1063/1.2335971 (1997)64:5%3C501::AID-QUA3%3E3.0.CO;2-%23
[19] A. Daneshvar, M. Moosavi and H. Sabzyan, Phys. Chem. [38] M. Tao, Phys. Rev. E 94, 043303 (2016). doi:10.1103/
Chem. Phys. 22, 13070–13083 (2020). doi:10.1039/ PhysRevE.94.043303
C9CP06994A [39] L. Monzel, A. Pausch, L.D.M. Peters, E.I. Tellgren, T. Hel-
[20] T. Detmer, P. Schmelcher, F.K. Diakonos and L.S. Ceder- gaker and W. Klopper, J. Chem. Phys. 157, 054106 (2022).
baum, Phys. Rev. A – At. Mol. Opt. Phys.56, 1825–1838 doi:10.1063/5.0097800
(1997). doi:10.1103/PhysRevA.56.1825 [40] T. Tajima, Computational Plasma Physics: With Applica-
[21] T. Detmer, P. Schmelcher and L.S. Cederbaum, Phys. tions to Fusion and Astrophysics (CRC Press, Boca Raton,
Rev. A – At. Mol. Opt. Phys. 57, 1767–1777 (1998). 2018).
doi:10.1103/PhysRevA.57.1767 [41] W. Magnus, Commun. Pure Appl. Math. 7, 649–673
[22] E.I. Tellgren, S.S. Reine and T. Helgaker, Phys. Chem. (1954). doi:10.1002/cpa.3160070404
Chem. Phys. 14, 9492–9499 (2012). doi:10.1039/ [42] S. Blanes, F. Casas, J.A. Oteo and J. Ros, Phys. Rep. 470,
c2cp40965h 151–238 (2009). doi:10.1016/j.physrep.2008.11.001
[23] K.K. Lange, E.I. Tellgren, M.R. Hoffmann and T. Helgaker, [43] A. Gómez Pueyo, M.A. Marques, A. Rubio and A. Cas-
Science (80-.). 337, 327–331 (2012). doi:10.1126/science. tro, J. Chem. Theory Comput. 14, 3040–3052 (2018).
1219703 doi:10.1021/acs.jctc.8b00197
[24] E.I. Tellgren, A.M. Teale, J.W. Furness, K.K. Lange, U. [44] M. Barbatti, G. Granucci, M. Persico, M. Rucken-
Ekström and T. Helgaker, J. Chem. Phys.140, 034101 bauer, M. Vazdar, M. Eckert-Maksić and H. Lischka,
(2014). doi:10.1063/1.4861427 J. Photochem. Photobiol. A Chem. 190, 228–240 (2007).
[25] S. Stopkowicz, J. Gauss, K.K. Lange, E.I. Tellgren doi:10.1016/j.jphotochem.2006.12.008
and T. Helgaker, J. Chem. Phys. 143, 074110 (2015). [45] A.I. Pausch, Ph. D. thesis, Karlsruher Institut für Tech-
doi:10.1063/1.4928056 nologie (KIT), 2022.
[26] J. Austad, A. Borgoo, E.I. Tellgren and T. Helgaker, [46] T.H. Dunning, J. Chem. Phys. 90, 1007–1023 (1989).
Phys. Chem. Chem. Phys. 22, 23502–23521 (2020). doi:10.1063/1.456153
doi:10.1039/D0CP03259J [47] R.A. Kendall, T.H. Dunning and R.J. Harrison, J. Chem.
[27] P. Schmelcher, L.S. Cederbaum and H.D. Meyer, Phys. Phys. 96, 6796–6806 (1992). doi:10.1063/1.462569
Rev. A 38, 6066–6079 (1988). doi:10.1103/PhysRevA.38. [48] D.E. Woon and T.H. Dunning, J. Chem. Phys. 100,
6066 2975–2988 (1994). doi:10.1063/1.466439
[28] D. Ceresoli, R. Marchetti and E. Tosatti, Phys. Rev. [49] E.I. Tellgren, A. Soncini and T. Helgaker, J. Chem. Phys.
B – Condens. Matter Mater. Phys. 75, 161101 (2007). 129, 154114 (2008). doi:10.1063/1.2996525
doi:10.1103/PhysRevB.75.161101 [50] E. Tellgren, T. Helgaker, A. Soncini, K.K. Lange, A.M.
[29] L.D.M. Peters, T. Culpitt, L. Monzel, E.I. Tellgren Teale, U. Ekström, S. Stopkowicz, J.H. Austad and S.
and T. Helgaker, J. Chem. Phys. 155, 024105 (2021). Sen, LONDON, a quantum-chemistry program for plane-
doi:10.1063/5.0056235 wave/GTO hybrid basis sets and finite au field calculations.