The LOTF Scheme for Biochemical Applications Steven Winfield , Gábor Csányi, James Kermode, Mike Payne Cavendish Laboratory, Cambridge Gianpietro Moras, Alessio Comisso, Alessandro De Vita King's College, London; University of Trieste Monika Fuxreiter, Ivan Solt Institute of Enzymology, Budapest Tristan Albaret Université Claude Bernard Lyon 1
20
Embed
The LOTF Scheme for Biochemical Applications Steven Winfield, Gábor Csányi… · 2008-09-06 · The LOTF Scheme for Biochemical Applications Steven Winfield, Gábor Csányi, James
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The LOTF Scheme for Biochemical Applications
Steven Winfield, Gábor Csányi, James Kermode, Mike PayneCavendish Laboratory, Cambridge
Gianpietro Moras, Alessio Comisso, Alessandro De VitaKing's College, London; University of Trieste
Monika Fuxreiter, Ivan SoltInstitute of Enzymology, Budapest
Tristan AlbaretUniversité Claude Bernard Lyon 1
Introduction
● First proposed in 1998 MRS proceedings● Validation in PRL paper in 2004● Recent change in formulation● Fundamental ingredients and goals unchanged● New scheme can work with any existing force model
(including a QM/MM model)
Outline:● Aim of LOTF● Previous method vs. new method● Benefits of the new formulation● Interfacing with existing QM/MM codes● Work in progress
Aim of LOTF
Two broad categories of problems when partitioning the system:
QM
CL
Long range: Interactions which decay slowly with distance, e.g. Coulomb:QM charge density ↔ CL point charges
`
Short range: Boundary cuts covalent bonds. Termination required for good electronic structure calculations
LOTF
What do we mean by good?1) verify thermodynamic properties in example systems2) correct FORCES for ALL QM ATOMS in a sample of configurations
Previous Learn on the Fly Method
● Take a classical potential for the system of interest
FMM
Previous Learn on the Fly Method
● Take a classical potential for the system of interest● Calculate accurate forces in the QM region
FQM
, FMM
Previous Learn on the Fly Method
● Take a classical potential for the system of interest● Calculate accurate forces in the QM region
FQM
HOW ?
Use a finite buffer region, throw away bad forces near surface
Previous Learn on the Fly Method
● Take a classical potential for the system of interest● Calculate accurate forces in the QM region● Tweak classical potential parameters to reproduce QM and MM forces. Carry out dynamics with this tweaked potential
F*MM
(α) ≈ {FQM
, FMM
}
New Learn on the Fly Method
● Using 'base' force model, calculate F0 on entire system
F*MM
(α) ≈ {FQM
, FMM
} F0 = {F
MM}
New Learn on the Fly Method
● Using 'base' force model, calculate F0 on entire system
● Calculate F1 : quantum mechanical forces in the QM region and take
classical forces in the MM region
F*MM
(α) ≈ {FQM
, FMM
} F1 = {F
QM, F
MM}
New Learn on the Fly Method
● Using 'base' force model, calculate F0 on entire system
● Calculate F1 : quantum mechanical forces in the QM region and take
classical forces in the MM region● Extend QM region
F*MM
(α) ≈ {FQM
, FMM
} F1 = {F
QM, F
MM}
New Learn on the Fly Method
● Create spline potential in extended region
F*MM
(α) ≈ {FQM
, FMM
}
New Learn on the Fly Method
● Create spline potential in extended region● Tune spline parameters α so that F
spline (α) ≈ F
1 - F
0
F*MM
(α) ≈ {FQM
, FMM
} Fspline
(α) ≈ F1 – F
0
New Learn on the Fly Method
● In fit region create spline potential● Tune spline parameters α so that F
spline (α) ≈ F
1 - F
0
● Carry out dynamics with FMM
+ Fspline
(α)
F*MM
(α) ≈ {FQM
, FMM
} FMM
+ Fspline
(α) ≈ {FQM
, FMM
}
Comparison of Methods
●Previous method: F*MM
(α) ≈ {FQM
, FMM
}
● Classical potential can have many parameters, e.g. bonds, angles, dihedrals… all require fitting.● Classical potential must be tightly integrated into LOTF code
●New method: FMM
+ Fspline
(α) ≈ {FQM
, FMM
}
● Concept of force fitting is kept, as is the use of buffer regions● Spline potential is simple to construct, easier to optimise● Details of classical potential do not enter into the method
LOTF Using Biochemical Force Models
FMM
+ Fspline
(α) ≈ {FQM
, FMM
}
FQM
and FMM
are calls to arbitrary force models, so there is no reason
why they cannot be replaced with:
F(1)QM/MM + F
spline(α) ≈ {F(2)
QM/MM, F(1)
QM/MM}
i.e. two calls to an existing QM/MM program, one with larger QM region, including buffer zone.
F(1)QM/MM
: “quick-to-compute” small quantum regions
F(2)QM/MM
: forces calculated with the required quantum region +
buffer zone to ensure QM force convergence. Forces on atoms in buffer zone are discarded.
Dynamics carried out by LOTF Dynamics carried out by LOTF →→ QMQM region can be region can be redefined at redefined at everyevery time step time step
Using LOTF
LOTFAtoms in Atoms out
User program:Quantum selection
QM/MM program
Using LOTF
LOTFAtoms in Atoms out
User program:Quantum selection
QM/MM program
Work In Progress
● Interface to AMBER 9● Constrained dynamics● Validation: g(r) of water for different models (pure AM1 and hybrid AM1 / TIP3P)
Conclusion
● New LOTF formulation
● Can use any existing QM/MM code
● Extends capability of QM/MM code
● Instantaneously correct QM forces / trajectory
● Movable QM region
Any questions or comments?
Thank-you for listening!
References:
[1] Gabor Csányi, T. Albaret, M. C. Payne, A. DeVita, '‘‘Learn on the Fly’’: A Hybrid Classical and Quantum-Mechanical Molecular Dynamics Simulation', Phys. Rev. Lett. 93(17) 175503 (2004)
[2] Gabor Csányi, T. Albaret, G. Moras, M. C. Payne, A. De Vita, 'Multiscale hybrid simulation methods for material systems' J. Phys. Cond. Matt. 17 R691-R703 (2005)