In molecular dynamics (MD) simulations temperature can be controlled via numerous methods. Out of those methods the most popular method is Nose-Hoover (NH) thermostat. The steps that are to be followed to implement NH equations in MD using Predictor Corrector algorithm is a bit challenging. So, I decided to lay an outline for the same. The steps are:
1. Prior to entering the main MD loop, evaluate forces [through a function called force( ) ]once and initialize the damping parameters (call it ${\eta}$) at 0.
2. In the main MD loop (let us say you want N simulations) call a function [let us say integrate( ) ] that evaluates the updated position and velocities
3. Within integrate( ):
3.1. Predictor Steps:
3.1.a) Evaluate force( ) and let us say is stored in a vector called ${F}$
3.1.b) Predict the updated positions through:
3.1.c) Predict the updated velocities through:
3.1.d) Predict the updated forces through:
3.1.e) Evaluate whether the predicted positions satisfy the periodic boundary conditions
3.2. Corrector Steps:
3.1.a) Evaluate force( ) let us say the force output is stored in vector ${F_{new}\left(t+h\right)}$
3.1.b) Update the new force to reflect the effect from bath:
3.1.c) Calculate the difference between the ${F_{new}}$ and ${F}$:
3.1.d) Correct the updated positions through:
3.1.e) Correct the updated velocities through:
3.1.f) Correct the updated forces through:
3.1.g) Correct the updated b through:
3.1.h) Check for periodic boundary conditions with the updated corrected positions
3.3. Update eta with the expression using NH equations of motion
4. Repeat step 3 till N simulations are complete.
NOTE: In Gear's algorithm: ${p = 1/12, q = 5/12, r = 1}$ and ${s = 1}$
PS: Please feel free to contact if you need any help or if you find any mistakes
1. Prior to entering the main MD loop, evaluate forces [through a function called force( ) ]once and initialize the damping parameters (call it ${\eta}$) at 0.
2. In the main MD loop (let us say you want N simulations) call a function [let us say integrate( ) ] that evaluates the updated position and velocities
3. Within integrate( ):
3.1. Predictor Steps:
3.1.a) Evaluate force( ) and let us say is stored in a vector called ${F}$
3.1.b) Predict the updated positions through:
${x\left(t+h\right) = x\left(t\right) + v\left(t\right)h + \dfrac{1}{2}a\left(t\right)h^2 + \dfrac{1}{6}b\left(t\right)h^3}$
${v\left(t+h\right) = v\left(t\right) + a\left(t\right)h + \dfrac{1}{2}b\left(t\right)h^2}$
${F\left(t+h\right) = F\left(t\right) + b\left(t\right)h}$
3.2. Corrector Steps:
3.1.a) Evaluate force( ) let us say the force output is stored in vector ${F_{new}\left(t+h\right)}$
3.1.b) Update the new force to reflect the effect from bath:
${F_{new}\left(t+h\right)=F_{new}\left(t+h\right)-\eta v\left( t+h \right)}$
${factor = F_{new}\left(t+h\right) - F\left(t+h\right)}$
${x_c\left(t+h\right) = x\left(t+h\right) + p \times factor h^2}$
${v_c\left(t+h\right) = v\left(t+h\right) + q \times factor h}$
${F_c\left(t+h\right) = F\left(t+h\right) + r \times factor}$
${b_c\left(t+h\right) = b\left(t+h\right) + \dfrac{s \times factor}{h}}$
3.3. Update eta with the expression using NH equations of motion
4. Repeat step 3 till N simulations are complete.
NOTE: In Gear's algorithm: ${p = 1/12, q = 5/12, r = 1}$ and ${s = 1}$
PS: Please feel free to contact if you need any help or if you find any mistakes
Which criteria do you believe should be used in choosing an integrator ?
ReplyDeleteIt is always advisable to proceed with a symplectic integrator. Please see the paper titled "Explicit reversible integrators for extended systems dynamics" for good symplectic integrators. Unfortunately, not all thermostatted dynamics have symplectic integrators. In such cases, predictor-corrector is a reasonably good alternative.
Delete