Boltzmann-conserving classical dynamics in quantum · PDF fileBoltzmann-conserving classical...
1
Boltzmann-conserving classical dynamics in quantum time-correlation functions: ‘Matsubara dynamics’ Michael J. Willatt , Timothy J. H. Hele and Stuart C. Althorpe Department of Chemistry, University of Cambridge, United Kingdom [email protected] Quantum Time-Correlation Functions These describe a whole host of dynamical properties, including I Chemical reaction rates (flux-side correlation functions) I Dipole absorption spectra (dipole moment autocorrelation functions) I Di↵usion constants (velocity autocorrelation functions). The quantum time-correlation function for the operators ˆ A and ˆ B is C AB (t )= Tr h e -β ˆ H ˆ Ae +i ˆ Ht /~ ˆ Be -i ˆ Ht /~ i and combines quantum Boltzmann statistics and quantum dynamics. I Most single-surface chemistry can likely be understood with a combination of quantum Boltzmann statistics and classical dynamics. I For example, an accurate description of liquid water involves lots of quantum statistics with very little quantum phase e↵ects. I Classical dynamics is much easier to compute than quantum dynamics. Linearised Semiclassical Initial Value Representation (LSC-IVR) Linearised Semiclassical Initial Value Representation (LSC-IVR) is the existing theory for combining quantum statistics and classical dynamics for quantum real-time correlation functions. I Truncation of the quantum Liouvillian to zeroth order in powers of ~ 2 gives the classical dynamics. I The quantum Boltzmann distribution is not conserved. I Detailed balance is not satisfied. I These shortcomings render LSC-IVR inadequate for a correct description of quantum statistics and classical dynamics. 0.5 0.55 0.6 0.65 0 2 4 6 8 10 D q 2 E t (a.u.) Figure 1 : Time-dependence of thermal expectation values in LSC-IVR. Exact quantum (black) and LSC-IVR (blue). Matsubara Dynamics We have discovered another theory which successfully combines quantum statistics and classical dynamics: Matsubara dynamics. 1 I Truncation of the quantum Liouvillian in normal mode derivatives, to the lowest Matsubara frequencies, gives the Matsubara Liouvillian, L M , (classical dynamics). I The quantum Boltzmann distribution is conserved. I Detailed balance is satisfied. The corresponding correlation function is C [M ] AB (t ) / Z d e P Z d e Q A( e Q)e -β [H M ( e P, e Q)+i ✓ M ( e P, e Q)] e L M t B ( e Q) where the Matsubara Hamiltonian is given by H M ( e P, e Q)= e P 2 2m + e U M ( e Q) and the potential, e U M ( e Q), is dependent upon M Matsubara modes only. Comparison of LSC-IVR and Matsubara Dynamics The distribution function in Matsubara dynamics contains a complex exponential of a phase ✓ M ( e P, e Q)= (M -1)/2 X n =-(M -1)/2 e P n e ! n e Q -n where e ! n =2⇡ n /β ~ is the n th Matsubara frequency and the summation is taken over all M of the normal modes which remain after truncation of the quantum Liouvillian. The classical Matsubara dynamics conserves this phase by virtue of maintaining smooth particle distributions in imaginary time. Figure 2 : Comparison of LSC-IVR evolution and Matsubara evolution for a two-dimensional quartic potential. LSC-IVR evolution leads to jagged imaginary time particle distributions which prevent conservation of the phase and the quantum Boltzmann distribution as a result. This often leads to poor agreement with quantum time-correlation functions. -0.4 -0.2 0 0.2 0.4 0 5 10 15 20 ˜ C qq (t) t (a.u.) Figure 3 : Comparison of LSC-IVR and Matsubara dynamics for a one-dimensional quartic potential. Exact quantum (black), LSC-IVR (blue) and Matsubara dynamics (red). Significance I The Matsubara dynamics time-correlation function is the starting point for any approximate method of treating quantum statistics and classical dynamics. I Existing approximate methods - LSC-IVR, Centroid Molecular Dynamics (CMD) and Ring Polymer Molecular Dynamics (RPMD) - can all be explained by Matsubara dynamics. I Matsubara dynamics may lead to other computationally tractable approximate methods. I Matsubara dynamics explains how quantum statistics and classical dynamics should be combined in quantum time-correlation functions. References [1] T. J. H. Hele, M. J. Willatt, A. Muolo and S. C. Althorpe, J. Chem. Phys. 142, 134103 (2015).
-
Upload
nguyenxuyen -
Category
Documents
-
view
227 -
download
5
Transcript of Boltzmann-conserving classical dynamics in quantum · PDF fileBoltzmann-conserving classical...
Boltzmann-conserving classical dynamics in quantum
time-correlation functions: ‘Matsubara dynamics’
Michael J. Willatt, Timothy J. H. Hele and Stuart C. Althorpe
Department of Chemistry, University of Cambridge, United [email protected]
Quantum Time-Correlation Functions
These describe a whole host of dynamical properties, including
IChemical reaction rates (flux-side correlation functions)
IDipole absorption spectra (dipole moment autocorrelation functions)
IDi↵usion constants (velocity autocorrelation functions).
The quantum time-correlation function for the operators A and B is
CAB(t) = Trhe��HAe+i Ht/~Be�i Ht/~
i
and combines quantum Boltzmann statistics and quantum dynamics.
IMost single-surface chemistry can likely be understood with a combinationof quantum Boltzmann statistics and classical dynamics.
IFor example, an accurate description of liquid water involves lots ofquantum statistics with very little quantum phase e↵ects.
IClassical dynamics is much easier to compute than quantum dynamics.
Linearised Semiclassical Initial Value Representation (LSC-IVR)
Linearised Semiclassical Initial Value Representation (LSC-IVR) is theexisting theory for combining quantum statistics and classical dynamics forquantum real-time correlation functions.
ITruncation of the quantum Liouvillian to zeroth order in powers of ~2 givesthe classical dynamics.
IThe quantum Boltzmann distribution is not conserved.
IDetailed balance is not satisfied.
IThese shortcomings render LSC-IVR inadequate for a correct description ofquantum statistics and classical dynamics.
0.5
0.55
0.6
0.65
0 2 4 6 8 10
D q2E
t (a.u.)Figure 1 : Time-dependence of thermal expectation values in LSC-IVR. Exactquantum (black) and LSC-IVR (blue).
Matsubara Dynamics
We have discovered another theory which successfully combines quantumstatistics and classical dynamics: Matsubara dynamics.1
ITruncation of the quantum Liouvillian in normal mode derivatives, to thelowest Matsubara frequencies, gives the Matsubara Liouvillian, LM,(classical dynamics).
IThe quantum Boltzmann distribution is conserved.
IDetailed balance is satisfied.
The corresponding correlation function is
C[M ]AB (t) /
Zd eP
Zd eQ A(eQ)e��[HM(eP,eQ)+i✓M(eP,eQ)]eLMtB(eQ)
where the Matsubara Hamiltonian is given by
HM(eP, eQ) =eP
2
2m+ eUM(eQ)
and the potential, eUM(eQ), is dependent upon M Matsubara modes only.
Comparison of LSC-IVR and Matsubara Dynamics
The distribution function in Matsubara dynamics contains a complexexponential of a phase
✓M(eP, eQ) =
(M�1)/2X
n=�(M�1)/2
ePne!neQ�n
where e!n = 2⇡n/�~ is the nth Matsubara frequency and the summation istaken over all M of the normal modes which remain after truncation of thequantum Liouvillian. The classical Matsubara dynamics conserves this phaseby virtue of maintaining smooth particle distributions in imaginary time.
Figure 2 : Comparison of LSC-IVR evolution and Matsubara evolution for atwo-dimensional quartic potential.
LSC-IVR evolution leads to jagged imaginary time particle distributionswhich prevent conservation of the phase and the quantum Boltzmanndistribution as a result. This often leads to poor agreement with quantumtime-correlation functions.
-0.4
-0.2
0
0.2
0.4
0 5 10 15 20
Cqq(t)
t (a.u.)Figure 3 : Comparison of LSC-IVR and Matsubara dynamics for a one-dimensionalquartic potential. Exact quantum (black), LSC-IVR (blue) and Matsubara dynamics(red).
Significance
IThe Matsubara dynamics time-correlation function is the starting point forany approximate method of treating quantum statistics and classicaldynamics.
IExisting approximate methods - LSC-IVR, Centroid Molecular Dynamics(CMD) and Ring Polymer Molecular Dynamics (RPMD) - can all beexplained by Matsubara dynamics.
IMatsubara dynamics may lead to other computationally tractableapproximate methods.
IMatsubara dynamics explains how quantum statistics and classicaldynamics should be combined in quantum time-correlation functions.
References
[1] T. J. H. Hele, M. J. Willatt, A. Muolo and S. C. Althorpe, J. Chem.Phys. 142, 134103 (2015).
