Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Introduction to 
Quantum Monte Carlo Methods 2! 
Claudio Attaccalite 
http://attaccalite.com
What we learned last time 
How to sample a given probability p(x) distribution with 
Metropolis Algorithm: 
Axx'=min1,...
Outline 
Path integral formulation of Quantum Mechanics 
Diffusion Monte Carlo 
One­Body 
density matrix and excitation en...
Path Integral : classical action 
The path followed by the particle is 
the one that minimize: 
t b dt Lxt , ˙xt ;t ...
Path Integral: Quantum Mechanics 
In quantum mechanics non just the 
extreme path contributes to the 
probability amplitud...
From Path Integral to Schrödinger equation 1 
X2 
X1 
X4 X5 
X3 
X... 
XM­1 
XA XB 
M m2 
SM=Σ j=1 
x j−x j−1 
−V x j...
From Path Integral to Schrödinger equation 2 
xi , t= 1 
Substituting the discretized action 
∞ expiℏ 
A∫−∞ 
m2 
...
Cafe Moment 
∞ dx0K x ,t ,x0,0 x0,0 
x ,t =∫−∞ 
What we have: ­> 
I=∫ f x1,. .., xNpx1,. .., xNdx1. ..dxN What...
Imaginary Time Evolution 
We want to solve the Schrödinger equation 
in imaginary time: ℏ ∂ 
=it 
∂  
= ℏ 
2m 
∂2 
∂ x...
From Path Integral to DMC: 1 
∞ dx0K x , , x0,0 x0,0 
 x ,=∫−∞ 
Using Feynman path integral 
the imaginary time ...
From Path Integral to DMC: 2 
K x, ,x0 ,0=lim 
N ∞ 
∞ Π j=1 
∫−∞ 
N−1 
dx j Σ 
N 
Wxn×Pxn , xn−1 x0,0 
n=1 
2...
The Algorithm 
We want generate the 
probability distribution 
N 
Pxn , xn−1 
Px0, x1,... xn , xN= x0,0Π i=1 
and s...
An example H and H2 
Convergence of the Energy 
H molecule versus  
H atom wave­function 
and energy
Application to Silicon: 
one body density matrix 
r , r '=Σi 
, j 
i , ji r j r ' i r  LDA local orbitals 
Th...
Results on Silicon 
Max difference between ii 
QMC and LDA is 0.00625 
Max off­diagonal 
element 
0.0014(1)
Results on Silicon: 2 
QMC one­body­density 
matrix 
on the 110 plane where r is fixed 
on the center of the bonding 
Diff...
Reference 
SISSA Lectures on Numerical methods for strongly correlated 
electrons 4th draft 
S. Sorella G. E. Santoro and ...
From Path Integral to Schrödinger equation: 1+1/2 
xi , t= 1 
Substituting the discretized action 
∞ expiℏ 
A∫−∞ 
...
Introduction to Diffusion Monte Carlo
Introduction to Diffusion Monte Carlo
Introduction to Diffusion Monte Carlo
Upcoming SlideShare
Loading in …5
×

Introduction to Diffusion Monte Carlo

881 views

Published on

Introduction to Diffusion Monte Carlo

Published in: Education
  • Be the first to comment

Introduction to Diffusion Monte Carlo

  1. 1. Introduction to Quantum Monte Carlo Methods 2! Claudio Attaccalite http://attaccalite.com
  2. 2. What we learned last time How to sample a given probability p(x) distribution with Metropolis Algorithm: Axx'=min1, px'T x, x' pxT x', x How to evaluate integrals in the form: Evaluate Quantum Mechanical Operators:  〈 f 〉=∫ f xpxdx= 1N Σ f xi where xi are distributed according to p(x) 〈 f  2 〉−〈 f 〉2  f=N 〈 A〉=∫ A dx ∫ x2 dx AL x= x  x =∫ AL xpxdx px= ∣x∣2 ∫∣x∣2dx
  3. 3. Outline Path integral formulation of Quantum Mechanics Diffusion Monte Carlo One­Body density matrix and excitation energies
  4. 4. Path Integral : classical action The path followed by the particle is the one that minimize: t b dt Lxt , ˙xt ;t  S=∫t a where S is the Classical Action and L is the Lagrangian Lxt  ,˙x t ;t=m2 ˙x t2−V xt  ;t  Only the extreme path contributes!!!!
  5. 5. Path Integral: Quantum Mechanics In quantum mechanics non just the extreme path contributes to the probability amplitude K B, A= Σ[xt] over allpossiblepaths PB, A=∣K 2,1∣2 where [xt]=Aexp{iℏS[ xt]} Feynman's path integral formula B expiℏ K B, A=∫A S[B, A]Dxt
  6. 6. From Path Integral to Schrödinger equation 1 X2 X1 X4 X5 X3 X... XM­1 XA XB M m2 SM=Σ j=1 x j−x j−1 −V x j2  2 It is possible to discretized the integral on the continuum into many intervals M slices of length =∣xi1−xi∣ ∞ K x2, t2 ; x1, t1 x1,t1  x2,t2=∫−∞ K B, A=lim ∞ ∫...∫expiℏ SM[2,1] dx1 A ... dxM−1 A  On each path the discretized classical action can be written as We want use this propagator in order to obtain the wave­function at time t2 in the position x2  xi , t= 1 ∞ expiℏ A∫−∞  , xi xi−1 , tdxi−1 Lxi−xi−1
  7. 7. From Path Integral to Schrödinger equation 2 xi , t= 1 Substituting the discretized action ∞ expiℏ A∫−∞ m2 exp[−i  ℏ V xi ,t ] xi , tdxi−1 We call xi−1−xi= , then send  , to zero and compare left and right at the same order  A=2 i ℏ  m 1/2 −ℏ i ∂ ∂t = −ℏ2 2m ∂2 ∂ x2V x ,t  At the order 0 we get the normalization constant At the order 1 we get the Schroedinger equation!
  8. 8. Cafe Moment ∞ dx0K x ,t ,x0,0 x0,0 x ,t =∫−∞ What we have: ­> I=∫ f x1,. .., xNpx1,. .., xNdx1. ..dxN What we want: ­>
  9. 9. Imaginary Time Evolution We want to solve the Schrödinger equation in imaginary time: ℏ ∂ =it ∂  = ℏ 2m ∂2 ∂ x2 [V x−Er ] The formal solution is: x , =exp[−H −ER ℏ ] x0 ,0 ∞ cnnxe− If we expand in a eigenfunction of H: x , =Σ n=0 En−ER ℏ if ER > E0 if ER < E0 if ER = E0 limt ∞   =∞ limt ∞  =o limt ∞  =0 Tree Possibility:
  10. 10. From Path Integral to DMC: 1 ∞ dx0K x , , x0,0 x0,0  x ,=∫−∞ Using Feynman path integral the imaginary time evolution can be rewritten as lim N∞ ∞ ...∫−∞ ∫−∞ ∞  m 2ℏ   N/2 exp{−  N [ m ℏ Σj =1 2  xi−x j−12V xi−En ]} K x, ,x0 ,0 is equal to and as usual we rewrite this integral as K x, ,x0 ,0=lim N ∞ ∞ Π j=1 ∫−∞ N−1 dx j Σ N Wxn×Pxn , xn1 x0,0 n=1
  11. 11. From Path Integral to DMC: 2 K x, ,x0 ,0=lim N ∞ ∞ Π j=1 ∫−∞ N−1 dx j Σ N Wxn×Pxn , xn−1 x0,0 n=1 2 ℏ exp[−mxn−xn−12 Pxn ,xn−1= m 2 ℏ  ] Wxn=exp[−[V xn −ER ]  2ℏ  ] A Gaussian probability distribution A Weight Function N Pxn , xn−1 Px0, x1,... xn , xN= x0,0Π i=1 N WxN f x1,... xn , xN=Π i=1 If we define: I=∫ f x1,. .., xN Px1,. .., xN dx1. ..dxN we have
  12. 12. The Algorithm We want generate the probability distribution N Pxn , xn−1 Px0, x1,... xn , xN= x0,0Π i=1 and sample N WxN f x1,... xn , xN=Π i=1 Generate points distributed on (x0,0) x1 is generate from x0 sampling P(xn,xn­1 ) (a Gaussian) the weight function is evaluated W(x1) x ,∞=0 X
  13. 13. An example H and H2 Convergence of the Energy H molecule versus  H atom wave­function and energy
  14. 14. Application to Silicon: one body density matrix r , r '=Σi , j i , ji r j r ' i r  LDA local orbitals The matrix elements are calculated as: i , j=N∫∗iri jr ' r ' ,r 2,.... , r N r1,. .., r N ∣r1,. .., r N2∣dr 'dr 1. ..dr N
  15. 15. Results on Silicon Max difference between ii QMC and LDA is 0.00625 Max off­diagonal element 0.0014(1)
  16. 16. Results on Silicon: 2 QMC one­body­density matrix on the 110 plane where r is fixed on the center of the bonding Difference between QMC and LDA for r=r' is 1.7%
  17. 17. Reference SISSA Lectures on Numerical methods for strongly correlated electrons 4th draft S. Sorella G. E. Santoro and F. Becca (2008) Introduction to Diffusion Monte Carlo Method I. Kostin, B. Faber and K. Schulten, physics/9702023v1 (1995) Quantum Monte Carlo calculations of the one­body density matrix and excitation energies of silicon P. R. C. Kent et al. Phys. Rev. B 57 15293 (1998) FreeScience.info­> Quantum Monte Carlo http://www.freescience.info/books.php?id=35
  18. 18. From Path Integral to Schrödinger equation: 1+1/2 xi , t= 1 Substituting the discretized action ∞ expiℏ A∫−∞  exp[−i mxi−xi−12 ℏ V xi ,t]xi−1 , tdxi−1 We call xi−1−xi= and send to zero ∂t xi , t= 1 A∫−∞ xi , t ∂ ∞ expim2 ℏ  [1−i ℏ V xi ,t...] [ xi ,t  ∂ ∂ xi  xi ,t 12 2  xi , t]dxi−1 2 ∂2 ∂ xi ∞ exp[imℏ2 1 A∫−∞ 2ℏ  ]d=1 and  ∂ ∂t =−i ℏ V − ℏ  2m ∂2 ∂ x2

×