Bulgac-Kusnezov-Nos-Hoover thermostats
Alessandro Sergi*
School of Physics, University of KwaZulu-Natal, Pietermaritzburg, Private Bag X01 Scottsville, 3209 Pietermaritzburg, South Africa
Gregory S. Ezra
Department of Chemistry and Chemical Biology, Baker Laboratory, Cornell University, Ithaca, New York 14853, USA Received 3 February 2010; published 19 March 2010 In this paper, we formulate Bulgac-Kusnezov constant temperature dynamics in phase space by means of non-Hamiltonian brackets. Two generalized versions of the dynamics are similarly dened, one where the Bulgac-Kusnezov demons are globally controlled by means of a single additional Nos variable, and another where each demon is coupled to an independent Nos-Hoover thermostat. Numerically stable and efcient measure-preserving time-reversible algorithms are derived in a systematic way for each case. The chaotic properties of the different phase space ows are numerically illustrated through the paradigmatic example of the one-dimensional harmonic oscillator. It is found that, while the simple Bulgac-Kusnezov thermostat is apparently not ergodic, both of the Nos-Hoover controlled dynamics sample the canonical distribution correctly. DOI: 10.1103/PhysRevE.81.036705 I. INTRODUCTION PACS number s : 05.10. a, 05.20.Jj, 05.45.Pq, 65.20.De
In condensed matter studies, there are many situations in which molecular dynamics simulation at constant temperature 13 is needed. For example, this occurs when magnetic systems are modeled in terms of classical spins 47 . Deterministic methods 810 , based on non-Hamiltonian dynamics 1119 , can sample the canonical distribution provided that the motion in the phase space of the relevant degrees of freedom is ergodic 1,3 . However, classical spin systems are usually formulated in terms of noncanonical variables 20,21 , without a kinetic energy expressed through momenta in phase space, so that Nos dynamics cannot be applied directly. To tackle this problem, Bulgac and Kusnezov BK introduced a deterministic constant-temperature dynamics 2224 , which can be applied to spins. A number of numerical approaches to integration of spin dynamics can be found in the literature 2528 . However, BK dynamics, as any other deterministic canonical phase space ow, is able to correctly sample the canonical distribution only if the motion in phase space is ergodic on the timescale of the simulation. In general, this condition is very difcult to check for statistical systems with many degrees of freedom, while it is known that, despite its simplicity, the one-dimensional harmonic oscillator provides a difcult and important challenge for deterministic thermostatting methods 9,2931 . In this paper, we accomplish two goals. First, by reformulating BK dynamics through non-Hamiltonian brackets 14,15 in phase space, we introduce two generalized versions of the BK time evolution which are able to sample the canonical distribution for a stiff harmonic system. Second, using a recently introduced approach based on the geometry of non-Hamiltonian phase space 19 , we are able to derive stable and efcient measure-preserving and time-reversible algorithms in a systematic way for all the phase space ows treated here.
The BK phase space ow introduces temperature control by means of ctitious coordinates and their associated momenta in an extended phase space traditionally called demons. Our generalizations of the BK dynamics are obtained by controlling the BK demons themselves by means of additional Nos-type variables 8 . In one case, the BK demons are controlled globally by means of a single additional NosHoover thermostat 8,9 . In the following this will be referred to as Bulgac-Kusnezov-Nos-Hoover BKNH dynamics. In the second case, each demon is coupled to an independent Nos-Hoover thermostat. This will be called the Bulgac-Kusnezov-Nos-Hoover chain BKNHC , and corresponds to massive NH thermostatting of the demon variables 32 . The ability to derive numerically stable measurepreserving time-reversible algorithms 19 for Nos controlled BK dynamics is very encouraging for future applications to thermostatted spin systems. This paper is organized as follows. In Sec. II, we briey sketch the unied formalism for non-Hamiltonian phase space ows and measure-preserving integration. The BK dynamics is formulated in phase space and a measurepreserving integration algorithm is derived in Sec. III. The BKNH and BKNH-chain thermostats are treated in Secs. IV and V, respectively. Numerical results for the onedimensional harmonic oscillator using these thermostats are presented and discussed in Sec. VI. Section VII reports our conclusions. In addition we include several appendices. A useful operator formula is derived in Appendix A, while invariant measures for the BK, BKNH, and BKNHC phase space ows are derived in Appendices BD, respectively.
II. NON-HAMILTONIAN BRACKETS AND MEASUREPRESERVING ALGORITHMS
*sergi@ukzn.ac.za
gse1@cornell.edu
Consider an arbitrary system admitting a timeindependent extended Hamiltonian expressed in terms of the phase space coordinates xi, i = 1 , . . . , 2N. In this case, the Hamiltonian can be interpreted as the conserved energy of the system.
2010 The American Physical Society
1539-3755/2010/81 3 /036705 14
036705-1
Upon introducing an antisymmetric tensor eld generalized Poisson tensor 21,33 in phase space, B x = BT x , one can dene non-Hamiltonian brackets 1416 as
2n
=
i,j=1
Bij H , xi x j
a,b =
i,j=1
a b Bij , xi xj
the statistical mechanics must be formulated in terms of a modied phase space measure 1218 = ew x where = dx1 dx2 . . . dx2N 9 8
where a = a x and b = b x are two arbitrary phase space functions. The bracket dened in Eq. 1 is classied as nonHamiltonian 1416 since, in general, it does not obey the Jacobi relation, i.e., in general the Jacobiator J 0, where 21 J = a, b,c + b, c,a + c, a,b , 2
is the standard phase space volume element volume form 35 and the statistical weight w x is dened by dw = dt x . 10
with c = c x arbitrary phase space function in addition to the functions a and b, previously introduced . If J 0, the tensor Bij is said to dene an almost-Poisson structure 34 such systems have also been called pseudo-Hamiltonian 33 . An energy conserving and in general non-Hamiltonian phase space ow is then dened by the vector eld
2N
xj is satised, then
xi = xi,H =
j=1
Bij
H , xj
L =0
for every
12
where conservation of H x follows directly from the antisymmetry of Bij. It has previously been shown how equilibrium statistical mechanics can be comprehensively formulated within this framework 16 . It is also possible to recast the above formalism and the corresponding statistical mechanics in the language of differential forms 17,18 . If the matrix B is invertible this is true for all the cases considered here , with inverse ij, we can dene the 2-form 35 1 = 2
ijdx i
so that the volume element is invariant under each of the L 19 . The condition 11 is satised for all the cases considered below, so that, exploiting the decomposition in Eq. 6 , algorithms derived by means of a symmetric Trotter factorization of the Liouville propagator:
ns1 ns1
exp L =
=1
exp
exp
=1
L ns 13
4
III. PHASE SPACE FORMULATION OF THE BK THERMOSTAT
The dynamics of Eq. 3 is then Hamiltonian if and only if the form Eq. 4 is closed, i.e., has zero exterior derivative, d = 0 35 . This condition is independent of the particular system of coordinates used to describe the dynamics. The structure of Eq. 3 can be taken as the starting point for derivation of efcient time-reversible integration algorithms that also preserve the appropriate measure in phase space 19 . Measure-preserving algorithms can be derived upon introducing a splitting of the Hamiltonian
ns
A phase space formulation of the BK thermostat can be achieved upon introducing the Hamiltonian HBK = K1 p p2 K2 p +V q + + + k BT 2m m m =H q,p + K1 p K2 p + + k BT m m + + 14a
14b
H=
=1
H ,
which in turn induces a splitting of the Liouville operator associated with the non-Hamiltonian bracket in Eq. 1 ,
2N
L xi = xi,H
=
j=1
Bij
H . xj
where q , p are the physical degrees of freedom coordinates and momenta , with mass m, to be simulated at constant temperature T, while and are the BK demons, with corresponding inertial parameters m and m , and associated momenta p , p 2224 . K1 and K2 provide the kinetic energy of demon variables, and for the moment are left arbitrary. Upon dening the phase space point as x = q , , , p , p , p = x1 , x2 , x3 , x4 , x5 , x6 , one can introduce an antisymmetric BK tensor eld as
036705-2
BULGAC-KUSNEZOV-NOS-HOOVER THERMOSTATS
0 0 0 BBK = 1 0 G2
0 0 0 0 G1 p 0
0 0 0 0 0 G2 q
1 0 0 0 G1 0
0 G1 p 0 G1 0 0
G2 0 G2 q 0 0 0
LBK = 1
V V + G2 , q p q p p p + G1 , m q m p G1 , p p G2 , q p G1 K1 , m p p .
18a
LBK = 2 , 15
18b
LBK = kBT 3
18c
LBK = kBT 4 1 G1 K1 m p p
18d
where G1 and G2 are functions of system variables p , q only. Substituting BBK and HBK into Eq. 3 , we obtain the energy-conserving equations q= H G2 q,p p m K2 , p 16a 16b 16c K1 , p 16d 16e 16f
LBK = 5
18e
LBK = 6
1 G2 K2 G2 K2 + m p q m q p
18f
= 1 G1 K1 , m p p = 1 G2 K2 , m q p p= H G1 q,p q m
Upon choosing a symmetric Trotter factorization of the BK Liouville operator based on the decomposition
8
BK
=
=1
LBK
19
p = G1 p = G2
G1 H , k BT p p G2 H . k BT q q
a measure-preserving algorithm can be produced in full generality. In practice, a choice of K1, K2, G1, and G2 must be made in order obtain explicit formulas. In this paper, we make the following simple choices: G1 = p, G2 = q, K1 = p2 , 2 p2 . 2 20a 20b
20c
In order to derive a measure preserving algorithms, the rst step, following Eq. 5 , is to introduce a splitting of HBK: HBK = V q , 1 HBK = 2 p2 , 2m 17a 17b 17c 17d 17e BBK = K2 =
20d
0 p
1 0
17f
Using the so-called direct translation technique 36 we can expand the above symmetric breakup of the Liouville operator into a pseudocode form, ready to be implemented on the computer: q i p q+ p 4m
23b
BK :UB
23c
p + Fp 4 p 2m p 2m
BK = k T L4 B BK = p L5 m
,
2
23d
p p exp q q exp
p p p + p m m +
23e
ii
p + 2m + p 2m p 4m
BK :UC
BK = p q + p L6 q m m
p2 . m p
23f
For the purposes of dening an efcient integration algorithm, we combine commuting Liouville operators as follows: iii
BK LA
q p
q+
BK :UB
BK + BK = F q L1 L4
+ Fp
24a
p + Fp 4 p+ F p + Fp q+ + p 4m p 4m
BK :UB
BK LB
BK + BK = p + Fp L2 L3 , p m q + p m
24b
iv
p p
BK :UA
BK LC
BK + BK = p p p q + p L5 L6 p m q m m
, q 24c v
p 25b
p + Fp 4 p 2m p 2m
Fp = Dening UBK
p2 kBT. m
BK :UC
= exp BK , L
= A , B , C, one possible reversible measurewhere preserving integration algorithm for the BK thermostat is then U
BK BK = UB
BK UC
BK UB
BK UA
q vii 27 p
q+
p 4m
BK :UB
BK UB
BK UC
BK UB
p + Fp 4
036705-4
The Hamiltonian can be split as HBKNH = V q , 1 HBKNH = 2 p2 , 2m 31a 31b 31c 31d 31e
p2 K1 p K2 p = H q,p + + + m m 2m + k BT + + 2kBT 28
is simply the BK Hamiltonian augmented by the Nos variables , p with mass m . With the antisymmetric BKNH tensor
BKNH
0 0 0 = 0 1 0 G2 0
0 0 0 0 0 G1 p 0 0
0 0 0 0 0 0 G2 q 0
0 0 0 0 0 0 0 1
1 0 0 0 0 G1 0 0
0 G1 p 0 0 G1 0 0 p
G2 0 G2 q 0 0 0 0 p
0 0
31f
0 1 0 p p 0 29 yields LBKNH = 1 ,
HBKNH = 7
31g 31h
HBKNH = 2kBT , 8
we obtain from Eq. 3 equations of motion for the phase space variables x = q , , , , p , p , p , p = x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 : q= H G2 q,p p m K2 , p 30a 30b
V V + G2 , q p q p p p + G1 , m q m p G1 , p p G2 , q p K1 , p p
33a
LBKNH = 2
33b
= 1 G1 K1 , m p p = 1 G2 K2 , m q p p = , m H G1 q,p p= q m K1 , p
LBKNH = kBT 3 30c LBKNH = kBT 4 30d 30e 30f LBKNH = 6 30g 30h LBKNH = 7 p m LBKNH = 5 1 G1 K1 m p p
33c
33d
p G1 K1 + m p p m
G1 H p p k BT , p = G1 p m p p = G2 p = p m G2 H p p k BT , q m q K1 p + p m K2 2kBT. p
Here, a single Nos variable is coupled to both of the BK demons and . The associated invariant measure is discussed in Appendix C.
036705-5
LBKNH = 2kBT 8
33h
At this stage, we leave the general formulation and adopt the particular choice of K1, K2, G1, and G2 given in Eq. 20 . The antisymmetric BKNH tensor becomes 0 0 0 BBKNH = 0 1 0 q 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 p 0 0 1 0 0 0 0 p q 0 1 0 0 0 0 p 0 0 0 1 0 p p 0 34 and the Hamiltonian simplies to
2 2 2 BKNH = H q,p + p + p + p + k T H B 2m 2m 2m
BKNH LB
BKNH LC
BKNH + BKNH = p + Fp L2 L3 p m q
37b
0 p
, where F q = V , q
38a
1 0
Fp = q + + 2 BT . 35
V kBT, q
38b
Fp =
p2 kBT, m
38c
The split Liouville operators are now V BKNH = V +q , L1 q p q p p2 BKNH = p + , L2 m q m p BKNH = k T L3 B , 36a
Fp =
p2 p2 + 2kBT. m m
38d
36b
36c
where k , i = , for LA. The action of the propagator associated with this operator on pi is derived in Appendix A, and is given by e Li p i = p ie
pk/mk
+ F pie
pk/2mk
BKNH = k T L4 B BKNH = p L5 m
pk 2mk
sinh
pk . 2mk 40
p2 p p + , p m p m + p2 , m p
sinh
pk 2mk
41
BKNH = p q + p L6 q m m BKNH = p L7 m
36f
p p p p , p p m m .
36g
is in fact well behaved as pk 0, and can be expanded in a Maclaurin series to suitably high order 37 . In our implementation we used an eighth order expansion. The propagators for the BKNH dynamics can now be dened as UBKNH = exp BKNH , L 42
BKNH = 2k T L8 B
36h
For the purposes of dening an efcient integration algorithm, we combine commuting Liouville operators as follows:
BKNH LA
= A , B , C. One possible reversible measurewhere preserving integration algorithm for the BKNH thermostat can then be derived from the following Trotter factorization: U
BKNH BKNH = UB
BKNH UC
BKNH UB
BKNH UA
BKNH UB
BKNH UC
BKNH UB
43
036705-6
BULGAC-KUSNEZOV-NOS-HOOVER THERMOSTATS
q i p p q ii
q+
p 4m
p
BKNH :UB
p exp q exp + + p 2m p 2m
p + Fp 4 p exp q exp + + p 2m p 2m p 2m p 2m
, q vi
p 2m p 2m
BKNH :UC
BKNH :UC
p + Fp 2 p 4m
q vii p
q+
p + Fp 2 q q+ p 4m
BKNH :UB
p + Fp 4
iii p p p iv p
BKNH :UB
p + Fp 4 p+ F q p + Fp + p m p m
V. BULGAC-KUSNEZOV-NOS-HOOVER CHAIN
BKNH :UA
For simplicity, we explicitly treat only the case, in which the p and p demons are each coupled to a standard NosHoover thermostat length one . It would be straightforward to couple each of the demons to NH chains 32 , and the general case can be easily inferred from what follows. Dene the Hamiltonian HBKNHC = H q,p + + k BT p2 p2 K1 p K2 p + + + m m 2m 2m + . 44
p exp p q+ 4m p + Fp 4
+ +
q v p
BKNH :UB
Upon dening the phase space point x = q , , , , , p , and the p , p , p , p = x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 , x9 , x10 antisymmetric BKNHC tensor
0 0 0 0 BBKNHC = 0 1 0 G2 0 0
0 0 0 0 0 0 G1 p 0 0 0
0 0 0 0 0 0 0 G2 q 0 0
0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 1
1 0 0 0 0 0 G1 0 0 0
0 G1 p 0 0 0 G1 0 0 p 0
G2 0 G2 q 0 0 0 0 0 0 p
0 0 0 1 0 0 p 0 0 0
0 0 0 0 1 0 0 p 0 0 , 45
036705-7
BBKNHC 0 46 0 0 0 = 0 1 0 q 0 47a 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 p 0 0 0 1 0 0 0 0 0 p 0 q 0 1 0 0 0 0 0 0 p 0 0 0 1 0 0 p 0 0 0 0 0 0 0 1 0 0 p 0 0 49 ,
xi =
BBKNHC ij
with i = 1 , . . . , 10.
Algorithm for BKNHC chain dynamics
0 p
1 0
HBKNHC = 2
47b
47c
+ k BT
+ +
50
and associated Liouville operators , 47d V BKNHC = V +q , L1 q p q p p2 BKNHC = p + , L2 m q m p BKNHC = k T L3 B 47f BKNHC = k T L4 B 47g BKNHC = p L5 m p , 51d 51e 51f 51g 51h 51i 51j p , 51a 51b 51c
HBKNHC = 5
K1 p , m
47e
K2 p , HBKNHC = 6 m
HBKNHC = 7
p2 , 2m
p2 p p + , p m p m + p2 , m p
47h
BKNHC = p q + p L6 q m m BKNHC = p L7 m
p p , p m p ,
47i
BKNHC = k T L8 B BKNHC = p L9 m
BKNHC = k BT , H10
47j
p p , p m p .
BKNHC = k T L10 B
BULGAC-KUSNEZOV-NOS-HOOVER THERMOSTATS
BKNHC LB
Li =
pk p i + F pi mk
pi
54
where k , i = , for LA and k , i = , for LB. Again following the derivation in Appendix A, we nd e Li p i = p ie
pk/mk
BKNHC LC
+ F pie
pk/2mk
pk 2mk
sinh
pk . 2mk 55
where F q = Fp = q Fp =
pk
pk 2mk
= exp BKNHC , L
56
V kBT, q
with = A , B , C can now be introduced. One possible reversible measure-preserving integration algorithm for the BKNHC chain thermostat is then U
BKNHC BKNHC = UB
p2 kBT, m
BKNHC UC
BKNHC UB
. 57
53e
and
BKNHC LB
q i p
q+ +
p 4m p 4m
/4 p /m BKNHC :UB
p e
+ F p e 4
/4 p /2m
p 4 2m
sinh
p 4 2m
p q
p exp q exp + p 2m
p 2m p 2m
ii p p
p + 2m
BKNHC :UC
p + Fp 2 p + Fp 2
036705-9
q iii p
q+ +
p 4m p 4m
/4 p /m BKNHC :UB
p e p
+ F p e 4
/4 p /2m
p 4 2m
sinh
p 4 2m
p+ F + p m
p /m BKNHC :UA
iv p
p e p 4m p 4m
+ F p e
p /2m
p 2m
sinh
p 2m
q v p
q+ +
BKNHC :UB
p e
/4 p /m
+ F p e 4 p q
/4 p /2m
p 4 2m
sinh
p 4 2m
p exp q exp + p 2m
p 2m p 2m
vi p p p 4m p 4m
/4 p /m
p + 2m
BKNHC :UC
p + Fp 2 p + Fp 2
q vii p
q+ +
BKNHC :UB
p e
+ F p e 4
/4 p /2m
p 4 2m
sinh
p 4 2m
In its simplicity, the dynamics of a harmonic mode in one dimension is a paradigmatic example for checking the chaotic ergodic properties of constant-temperature phase space ows and the correct sampling of the canonical distribution. It is well known that it is necessary to generalize basic Nos-
Hoover dynamics 1,8,9 to thermostats such as the NosHoover chain 32,37 in order to produce correct constanttemperature averages for systems such as the harmonic oscillator. Some time ago, BK dynamics was devised to provide a deterministic thermostat for systems such as classical spins
036705-10
BULGAC-KUSNEZOV-NOS-HOOVER THERMOSTATS
0.08
1.04 1.03 H
BKNH
BKNHC
4 3 2 1 0 -1 -2 -3 -4 -4-3-2-1 0 1 2 3 4 q
BK
FIG. 1. Comparison of the total extended Hamiltonian versus time normalized with respect to its value at t = 0 for the harmonic oscillator undergoing simple Bulgac-Kusnezov dynamics HBK , NH controlled Bulgac-Kusnezov dynamics HBKNH , and BulgacKusnezov-Nos-Hoover chain dynamics HBKNHC . Two curves have been displaced vertically for clarity. The time-reversible measure-preserving algorithms developed in this paper conserve the extended Hamiltonian to high accuracy in all three cases.
FIG. 2. Radial phase space probability for a harmonic oscillator under Bulgac-Kusnezov dynamics. The continuous line shows the theoretical value while the black bullets display the numerical results. The inset displays a plot of the phase space distribution of points along the single trajectory used to compute the radial probability.
23,24 . To ensure efcient thermostatting, BK found it necessary to introduce several demons per thermostatted degree of freedom, where each demon was taken to have a different and in principle complicated coupling to the system degree of freedom 23,24 . In the present work, we keep the form of the system-thermostat coupling as simple as possible, in order to facilitate the formulation of explicit, reversible and measure-preserving integrators 19 . It is then of interest to investigate the ability of our BK-type thermostats to produce the correct canonical sampling in the case of the harmonic oscillator. Interest in harmonic modes is also justied by the possibility of devising models of condensed matter systems in terms of coupled spins and harmonic modes, as already done in quantum dynamics with so-called spin-boson models 38 . We therefore investigate the performance of our integration schemes on the simple one-dimensional harmonic oscillator. For the particular calculations reported here, the oscillator angular frequency, all masses and kBT were taken to be unity. The time step in all cases was = 0.0025, and all runs were calculated for 106 time steps, starting from the same initial conditions: harmonic oscillator coordinate q = 0.3, all other phase space variables zero at t = 0. The measure-preserving algorithms derived here result in stable numerical integration for all the three cases treated: BK, BKNH, and BKNHC chain dynamics. Figure 1 shows the three extended Hamiltonians normalized by their respective initial time value versus time. All three Hamiltonians are numerically conserved by the corresponding measurepreserving algorithm to very high accuracy which is maintained in all the three cases . However, the basic BK phase space ow is not capable of producing the correct canonical sampling for a harmonic mode. This can be easily checked since the canonical distribution function of the harmonic oscillator is isotropic in phase space and its radial dependence can be calculated exactly. Details of this way of visualizing the phase space sampling have already been given in 14,15 . Figure 2, displaying the comparison between the theoretical and the calculated value of the radial probability in phase space,
clearly shows that the BK dynamics is not able to produce canonical sampling. A look at the inset of Fig. 2, showing the phase space distribution for the harmonic mode, also immediately shows that the dynamics is not ergodic. The same analysis has been carried out for BKNH and BKNHC phase space ows, and these are displayed in Figs. 3 and 4, respectively. Within numerical errors, both BKNH and BKNHC thermostats are able to produce the correct canonical distribution for the stiff harmonic modes. Introduction of a single, global Nos-type variable in the BKNH thermostat effectively introduces additional coupling between the two demon variables. The effectiveness of the BKNH thermostat is consistent with our ndings results not discussed here that introduction of explicit coupling between demons in BK thermostat dynamics also leads to efcient thermostatting of the harmonic oscillator.
VII. CONCLUSIONS
We have formulated Bulgac-Kusnezov 23,24 , NosHoover controlled Bulgac-Kusnezov, and Bulgac-KusnezovNos-Hoover chain thermostats in phase space by means of
0.08 0.06 0.04 0.02 0 0 0.5 1 1.5 2 2.5 3 3.5 4 r P
p 4 3 2 1 0 -1 -2 -3 -4 -4-3-2-1 0 1 2 3 4 q
FIG. 3. Radial phase space probability for a harmonic oscillator under Nos-Hoover controlled Bulgac-Kusnezov dynamics. The continuous line shows the theoretical value while the black bullets display the numerical results. The inset displays a plot of the phase space distribution of points along the single trajectory used to compute the radial probability.
036705-11
4 3 2 1 0 -1 -2 -3 -4 -4-3-2-1 0 1 2 3 4 q
pi
exp
pk p i + F pi mk
pi
pi ,
A3a
=pie
pk/mk
mk F p 1 e pk i
pk/mk
A3b
sinh =pie
pk/mk
+ F pie
pk/2mk
FIG. 4. Radial phase space probability for a harmonic oscillator under Bulgac-Kusnezov-Nos-Hoover chain dynamics. The continuous line shows the theoretical value while the black bullets display the numerical results. The inset displays a plot of the phase space distribution of points along the single trajectory used to compute the radial probability.
pk 2mk . pk 2mk
A3c
non-Hamiltonian brackets 14,15 . We have derived timereversible measure-preserving algorithms 19 for these three cases and showed that additional control by a single NosHoover thermostat or independent Nos-Hoover thermostats is necessary to produce the correct canonical distribution for a stiff harmonic mode. Measure-preserving dynamics of the kind discussed here is associated with equilibrium simulations where, for example, there is a single temperature parameter T . Stationary phase space distributions associated with nonequilibrium situations are much more complicated than the smooth equilibrium densities analyzed in the present paper 11,39,40 . Nonequilibrium simulations of heat ow could be carried out by extending the present approach to multimode systems e.g., a chain of oscillators coupled to BK-type demons with associated NH thermostats corresponding to two different temperatures 4143 . The techniques presented here for derivation and implementation of thermostats have been shown to be efcient and versatile. We anticipate that analogous approaches can be usefully applied to systems of classical spins coupled to both harmonic and anharmonic modes.
BBK HBK 1 G1 K1 1 G2 K2 ij = . xi xi m p p m q p B1
K1 K2 + , m m
B2
B3
dt
BK
B4a
B4b + , B4c
We wish to determine the action of the propagator associated with the Liouville operator Eq. 39 . This is equivalent to solving the evolution equation recall i k pk dpi = p i + F pi , dt mk from t = 0 to t = . Integrating, we have mk pk ln pi + F pi pk mk giving
The phase space compressibility of the NH controlled Bulgac-Kusnezov thermostat is BBKNH HBKNH ij BKNH = xi xi = 1 G1 K1 1 G2 K2 p 2 . m p p m q p m C1
A1
A2
BKNH HT =H+
p2 K1 K2 + + , m m 2m
C2
we have
036705-12
BULGAC-KUSNEZOV-NOS-HOOVER THERMOSTATS
BK 1 dHT , kBT dt
BKNH =
C3
BKNHC =
BBKNHC HBKNHC ij xi xi 1 G1 K1 1 G2 K2 p p . m p p m q p m m D1
d = dx exp
t
dt
BKNH
C4a
p2 p2 K1 K2 + + + , m m 2m 2m
BK 1 dHT , kBT dt
D2
C4b
we have
BKNHC =
D3
+ +2
C4c
1 S. Nos, Prog. Theor. Phys. 103, 1 1991 . 2 D. Frenkel and B. Smit, Understanding Molecular Simulation: From Algorithms to Applications, 2nd ed. Academic, New York, 2001 . 3 B. Leimkuhler and S. Reich, Simulating Hamiltonian Dynamics Cambridge University Press, Cambridge, 2004 . 4 K. Chen and D. P. Landau, Phys. Rev. B 49, 3266 1994 . 5 A. Bunker, K. Chen, and D. P. Landau, Phys. Rev. B 54, 9259 1996 . 6 H. G. Evertz and D. P. Landau, Phys. Rev. B 54, 12302 1996 . 7 B. V. Costa, J. E. R. Costa, and D. P. Landau, J. Appl. Phys. 81, 5746 1997 . 8 S. Nos, J. Chem. Phys. 81, 511 1984 . 9 W. G. Hoover, Phys. Rev. A 31, 1695 1985 . 10 J. Jellinek, J. Phys. Chem. 92, 3163 1988 . 11 D. J. Evans and G. P. Morriss, Statistical Mechanics of Nonequilibrium Liquids Academic, New York, 1990 . 12 M. E. Tuckerman, C. J. Mundy, and G. J. Martyna, Europhys. Lett. 45, 149 1999 . 13 M. E. Tuckerman, Y. Liu, G. Ciccotti, and G. J. Martyna, J. Chem. Phys. 115, 1678 2001 . 14 A. Sergi and M. Ferrario, Phys. Rev. E 64, 056125 2001 . 15 A. Sergi, Phys. Rev. E 67, 021101 2003 . 16 A. Sergi and P. V. Giaquinta, J. Stat. Mech. 2007, P02013 2007 . 17 G. S. Ezra, J. Math. Chem. 32, 339 2002 . 18 G. S. Ezra, J. Math. Chem. 35, 29 2004 . 19 G. S. Ezra, J. Chem. Phys. 125, 034104 2006 . 20 A. Bulgac and D. Kusnezov, Ann. Phys. 199, 187 1990 . 21 J. E. Marsden and T. S. Ratiu, Introduction to Mechanics and
Symmetry Springer-Verlag, New York, 1999 . 22 A. Bulgac and D. Kusnezov, Phys. Rev. A 42, 5045 1990 . 23 D. Kusnezov, A. Bulgac, and W. Bauer, Ann. Phys. 204, 155 1990 . 24 D. Kusnezov and A. Bulgac, Ann. Phys. 214, 180 1992 . 25 J. Frank, W. Z. Huang, and B. Leimkuhler, J. Comput. Phys. 133, 160 1997 . 26 T. Arponen and B. Leimkuhler, BIT Numerical Math. 44, 403 2004 . 27 S. H. Tsai, M. Krech, and D. P. Landau, Braz. J. Phys. 34, 384 2004 . 28 R. I. McLachlan and D. R. J. ONeale, J. Phys. A 39, L447 2006 . 29 W. G. Hoover, C. G. Hoover, and D. J. Isbister, Phys. Rev. E 63, 026209 2001 . 30 F. Legoll, M. Luskin, and R. Moeckel, Arch. Ration. Mech. Anal. 184, 449 2007 . 31 F. Legoll, M. Luskin, and R. Moeckel, Nonlinearity 22, 1673 2009 . 32 G. J. Martyna, M. L. Klein, and M. E. Tuckerman, J. Chem. Phys. 97, 2635 1992 . 33 J. D. Ramshaw, J. Non-Equilib. Thermodyn. 16, 33 1991 . 34 A. C. da Silva and A. Weinstein, Geometric Models for Noncommutative Algebra AMS, New York, 1999 . 35 B. Schutz, Geometrical Methods of Mathematical Physics Cambridge University Press, Cambridge, 1980 . 36 M. Tuckerman, B. J. Berne, and G. J. Martyna, J. Chem. Phys. 97, 1990 1992 . 37 G. J. Martyna, M. E. Tuckerman, D. J. Tobias, and M. L. Klein, Mol. Phys. 87, 1117 1996 . 38 A. J. Leggett, S. Chakravarty, A. T. Dorsey, M. P. A. Fisher,
036705-13
ALESSANDRO SERGI AND GREGORY S. EZRA A. Garg, and W. Zwerger, Rev. Mod. Phys. 59, 1 1987 . 39 W. G. Hoover, J. Chem. Phys. 109, 4164 1998 . 40 J. R. Dorfman, An Introduction to Chaos in Nonequilibrium Statistical Mechanics Cambridge University Press, Cambridge, 1999 . 41 C. J. Mundy, S. Balasubramanian, K. Bagchi, M. E. Tucker-
PHYSICAL REVIEW E 81, 036705 2010 man, G. J. Martyna, and M. L. Klein, Nonequilibrium Molecular Dynamics Wiley-VCH, New York, 2000 , Vol. 14, pp. 291397. 42 W. G. Hoover, Mol. Simul. 33, 13 2007 . 43 W. G. Hoover and C. G. Hoover, J. Chem. Phys. 126, 164113 2007 .
036705-14