# Register now to access 7 million high quality study materials (What's Course Hero?) Course Hero is the premier provider of high quality online educational resources. With millions of study documents, online tutors, digital flashcards and free courseware, Course Hero is helping students learn more efficiently and effectively. Whether you're interested in exploring new subjects or mastering key topics for your next exam, Course Hero has the tools you need to achieve your goals.

17 Pages

### 05QD

Course: CS 596, Fall 2008
School: USC
Rating:

Word Count: 4050

#### Document Preview

Dynamics Quantum Basics In this chapter, we will simulate the dynamics of a particle, such as an electron, which follows the law of quantum mechanics [1]. Basics of the quantum-dynamics (QD) method [2-5] are described, along with corresponding data structures in program, qd.c. Schrdinger Wave Equation WAVE FUNCTION The state of an electron at time t is specified by a complex-valued wave function, ( r ,t) = Re(...

Register Now

#### Unformatted Document Excerpt

Coursehero >> California >> USC >> CS 596

Course Hero has millions of student submitted documents similar to the one
below including study guides, practice problems, reference materials, practice exams, textbook help and tutor support.

Course Hero has millions of student submitted documents similar to the one below including study guides, practice problems, reference materials, practice exams, textbook help and tutor support.
Dynamics Quantum Basics In this chapter, we will simulate the dynamics of a particle, such as an electron, which follows the law of quantum mechanics [1]. Basics of the quantum-dynamics (QD) method [2-5] are described, along with corresponding data structures in program, qd.c. Schrdinger Wave Equation WAVE FUNCTION The state of an electron at time t is specified by a complex-valued wave function, ( r ,t) = Re( r ,t) + iIm( r ,t) C (where i = 1), which is spread in the 3-dimensional space, r = (x, y,z) R3. Given the wave function, we can calculate various physical properties such as: 2 2 2 P( r ,t) = * ( r ,t)( r ,t) = ( r ,t) = Re( r ,t) + Im( r ,t) : The probability to find the electron at position r at time t. 2 r (t) = dx dy dz ( r ,t) r : The expected position of the electron at time t. Here, * ( r ,t) = Re( r ,t) iIm( r ,t) is the complex conjugate of the wave function. Normalization: The electron wave function must be normalized such that the electron must be found somewhere in the entire space with probability 1, i.e., 2 (1) dx dy dz ( r ,t) = 1. WAVE EQUATION The time evolution of the electron state, subjected to a time-independent, real-valued potential, V ( r ), is described by the following partial differential equation: i 2 ( r ,t) = 2 + V ( r )( r ,t), t 2m (2) where = 1.0545710-27 gcm2/s is the Planck constant, m = 9.1093810-28 g is the electron mass, and 2 2 2 2 = 2 + 2 + 2 is the Laplacian operator. x y z Dimensionless equation: In the following, we measure length (x, y, z) in unit of 2 /me 2 = 0.52917710-8 cm, time (t) in unit of 3 /me 4 = 2.4188910-17 s, and energy (V) in unit of me 4 / 2 = 4.3597410-11 gcm2/s2, where e = 4.8032110-10 esu is the electron charge in the CGS unit. Substituting 2 2 2 x= x , y = y ,z = z me 2 me 2 me 2 3 t (3) t = 4 me me 4 V = 2 V in Eq. (2), we obtain 1 2 2 2 4 me 4 me 2 me i 3 ( r , t ) = + 2 V ( r ) ( r , t ), 2m 2 t or the dimensionless equation, 2 ( r , t ) = + V ( r )( r , t ). t 2 i (4) In the following, we will use the dimensionless variables discussed above but omit the symbol for brevity. TWO-DIMENSIONAL ELECTRON As a specific example, the program qd.c simulates the time evolution of an electron confined in the 2-dimensional plane (z = 0). Such electrons are common at the interface between two heterogeneous materials in semiconductor devices. The electron state is now specified by the 2-dimensional wave function, (x, y,t) , where 0 x Lx and 0 y Ly (Lx and Ly are the system sizes in the x and y directions, respectively), and its time evolution is governed by the 2-dimensional Schrdinger equation, In Eq. (5), the Hamiltonian operator, H, is defined as i (x, y,t) = H(x, y,t). t 1 2 1 2 + V (x, y) 2 x 2 2 y 2 . (5) H = (6) = Tx + Ty + V Boundary condition: We impose the periodic boundary condition on the wave function such that (x + L x , y) = (x, y) . (x, y + Ly ) = (x, y) (7) DISCRETIZATION The wave function is discretized on a regular mesh of size x and y in the x and y directions, respectively. Here x = Lx/Nx y = Ly/Ny, where Nx and Ny are the numbers of mesh points in the x and and y directions, respectively. We denote the discretized wave function as j,k = (jx, ky), see the figure below. Figure: 2-dimensional mesh, on which the wave function is discretized. 2 In the discretized form, the Hamiltonian operator in Eq. (6) acts as (H ) j,k = (Tx ) j,k + (Ty ) j,k + (V ) j,k , where (8) 1 j1,k 2 j,k + j +1,k (Tx ) j,k = 2 (x) 2 1 j,k1 2 j,k + j,k +1 = Ty j,k 2 (y) 2 (V ) j,k = V j,k j,k () (9) and the potential-energy function is discretized as Vj,k = V(jx, ky). Note that the discretized H is a mapping from an NxNy array (j,k is its element at the j-th column and k-th row) to another NxNy array H. The (j, k)-th element, (H)j,k, of the output array, H, is a linear combination of the input array with different indices, as specified in Eqs. (8) and (9). Numerical Integration of Schrdinger Equation The time evolution of the wave function is formally written down as (t + t) = exp(iHt )(t) , (10) where we omit the indices for simplicity. Here the exponential function of an operator is defined as a series expansion, 1 n exp(iHt ) = (iHt ) . (11) n=0 n! We can show that Eq. (10) is the solution of the Schrdinger equation, Eq. (5), as follows: (t + t) (t) i = i lim t0 t t exp(iHt )(t) (t) = i lim t0 t 1 2 1 iHt + (iHt) + (t) (t) 2 = i lim t0 t = i lim iHt(t) + O (t) 2 t ( ) t0 = H(t) We can also show that Eq. (10) satisfies the correct initial condition at t = 0: (t + 0) = exp(iH 0)(t) = 1(t) = (t) . Since Eq. (10) satisfies both the first-order differential equation and its initial condition, it is the solution of Eq. (5). // In the split-operator method [2-5], the wave function is propagated for a small time interval, t, as 3 (t + t) = exp(iVt /2) exp(iTx t ) exp(iTy t ) exp(iVt /2)(t) +O [t ] ( 3 ) () . (12) Note that, for operators (or for matrices) A and B, exp(A+B) exp(A)exp(B) in general. However, if both operators are multiplied by a small factor , the following symmetric splitting is a very accurate approximation: (13) exp((A + B)) = exp(A /2) exp(B) exp(A /2) + O 3 . This can be easily proven by expanding both sides in and comparing the corresponding terms of the same power of : 2 exp((A + B)) = 1+ (A + B) + (A 2 + AB + BA + B 2 ) + O 3 2 () A 2 A 2 A 2 A 2 2 B 2 exp(A /2) exp(B) exp(A /2) = 1+ + + 1+ B + + 1+ + + 2 8 2 2 8 A 2 B 2 A 2 AB BA A 2 A A = 1+ + B + + 2 + + + + + + O 3 2 2 8 2 8 2 2 4 () Equation (12) is obtained by repeatedly applying this symmetric splitting operation. // In Eq. (12), the application of the potential propagator, exp(-iVt/2), is straightforward. (exp(iVt /2) ) j,k = j,k = j,k it (it /2) 2 (V ) j,k + 2! V 2 j,k + 2 () 2 it (it /2) 2 V j,k j,k + V j,k j,k + . 2 2! = exp iV j,k t /2 j,k () (14) ( ) Note that, for real number a, exp(ia) = 1+ ia + 1 1 1 1 (a 2 ) + (ia 3 ) + (a 4 ) + (ia 5 ) + 2! 3! 4! 4! 1 1 1 1 = 1 a 2 + a 4 + + i a a 3 + a 5 + . 2! 4! 3! 5! = cos(a) + isin(a) (15) Using Eq. (15) in Eq. (14), (exp(iVt /2) ) j,k = [cos(V j,kt /2) isin(V j,kt /2)][Re j,k + iIm j,k ] = cos V j,k t /2 Re j,k + sin V j,k t /2 Im j,k ) ( ) [( ] +i[cos(V j,k t /2) Im j,k sin(V j,k t /2) Re j,k ] . (16) Many algorithms have been proposed to apply the kinetic propagators such as exp(-iTxt). Among these algorithms, the space-splitting method (SSM) [4,5] is highly scalable on massively parallel computers. To understand the SSM, we first note that the operation of Tx on j,k is expressed as Tx j,k = b j1,k + 2a j,k + b j +1,k (17) 4 where a = 1/2 x 2 () b = 1/2(x ) 2 (18) For each index k, the operation of Tx on j,k thus amounts to the multiplication of a tridiagonal matrix, 2a b b 2a b b 2a b Tx = . b 2a b b 2a b b 2a In the SSM, this tridiagonal matrix is expressed as a direct sum of 2 2 submatrices, 2a b Tx = a b 1 = 2 b 2a b b 2a b b 2a b b 2a b b 2a a a b b a ab b a ba ab ab 1 ba ba + + 2 ab b a ba b a b a a b a (19) (20) where we have omitted the index k. The exponential of a 22 matrix on the right-hand side of Eq. (20) is calculated analytically as follows: (half) (full) (half) exp(itTx ) = U x U x U x + O (t) 3 = ( ) + + + 2 2 2 1 2 + + + 1 1 2 2 2 2 + + + 2 2 1 1 2 2 (21) + + 2 2 2 2 + 1 1 + + + 2 1 1 2 2 2 + + + 2 1 2 2 2 5 where + 1 it it (a + b) + exp (a b) n = exp n 2 n it it = 1 exp (a + b) exp (a b) n 2 n n (22) The operation of exp(-iTyt) is executed in a similar manner. Equation (22) can be derived by utilizing the exponential of an 22 matrix, a b , b a which can be calculated in terms of its eigenvalues, . The eigenvalues are the nontrivial solutions of the eigenvalue equation, a bu u (23) = , b av v or equivalently a b u 0 (24) = , b av 0 where u v is the corresponding eigenvector. In fact, Eq. (23) has a trivial solution, u = v = 0, for general values of , for which the determinant, a b D= = ( a) 2 b 2 , (25) b a takes a non-zero value. In such a case, Eq. (24) can be inverted as u a b 10 a b 0 0 1 = = = . a0 0 v b a 0 ( a) 2 b 2 b The nontrivial solutions, i.e., the eigenvalues, on the other hand, make the determinant D = ( - a)2 - b2 = 0. Let us denote the two solutions of this equation, D = 0, as = a b. The corresponding eigenvectors obtained by examining Eq. (24): are + a b u+ b bu+ 0 u+ 1/ 2 = = = b + av + b b v + 0 v + 1/ 2 . b u b bu 0 u 1/ 2 a = = = b av b bv 0 v 1/ 2 (26) In Eq. (26), we have normalized the eigenvectors such that their norm is 1. Or the eigenvalue equation, Eq. (23), for these specific values reads 6 a b a b 1/ 2 b1/ 2 = (a + b) a1/ 2 1/ 2 . 1/ 2 b 1/ 2 = (a b) a1/ 2 1/ 2 (27) The two matrix equations in Eq. (27) can be combined as a b1/ 2 1/ 2 1/ 2 1/ 2 a + b 0 = , a b b a1/ 2 1/ 2 1/ 2 1/ 2 0 (28) or equivalently a b 1/ = b a 1/ 1/ = 1/ 0 1/ 2 1/ 2 a + b a b1/ 2 1/ 2 0 0 1/ 2 1/ 2 a + b a b1/ 2 1/ 2 0 1 2 1/ 2 2 1/ 2 2 1/ 2 . 2 1/ 2 (29) = UEU 1 Using Eq. (23), we can show that (a + b) n a bn n = UEU 1 = UEU 1UEU 1 UEU 1 = UE nU 1 = U b a 0 ( ) 0 1 U . (a b) n (30) By using Eq. (30), we can now calculate the exponential of the 22 matrix as follows: n a b n a bn n (a + b) n 0 1 exp U U = = (a b) n b a n=0 n! b a n=0 n! 0 n n (a + b) n 0 1 = U U n=0 n! 0 (a b) n exp( (a + b)) 1 0 = U . U 0 exp( (a b)) 1/ 2 1/ 2 exp( (a + b)) 1/ 2 1/ 2 0 = 0 exp( (a b))1/ 2 1/ 2 1/ 2 1/ 2 exp( (a + b)) + exp( (a b)) exp( (a + b)) exp( (a b)) 2 2 = exp( (a + b)) exp( (a b)) exp( (a + b)) + exp( (a b)) 2 2 (31) Equation (21) is a straightforward application of the formula, Eq. (31). // 7 Data Structures of qd.c NX, NY: Number of mesh points in the x and y directions. psi[NX+2][NY+2][2]: psi[i][j][0|1] is the real|imaginary part of the wave function on mesh point (i, j) in the xy plane. The wave function to be simulated is in the range, 1 i NX and 1 j NY. To simplify the operation of the finite-difference operators considering the periodic boundary condition, the wave function values at the edge are duplicated as follows: for (sy=1; sy<=NY; sy++) for (s=0; s<=1; s++) { psi[0][sy][s] = psi[NX][sy][s]; psi[NX+1][sy][s] = psi[1][sy][s]; } for (sx=1; sx<=NX; sx++) for (s=0; s<=1; s++) { psi[sx][0][s] = psi[sx][NY][s]; psi[sx][NY+1][s] = psi[sx][1][s]; } v[NX+2][NY+2]: v[i][j] is the potential energy at mesh point (i, j). u[NX+2][NY+2][2]: u[i][j][0|1] is the real|imaginary part of the potential propagator at mesh point (i, j). The potential propagator, exp(-iVt/2), is operated in qd.c as follows, see Eq. (16). for (sx=1; sx<=NX; sx++) for (sy=1; sy<=NY; sy++) { wr=u[sx][sy][0]*psi[sx][sy][0]-u[sx][sy][1]*psi[sx][sy][1]; wi=u[sx][sy][0]*psi[sx][sy][1]+u[sx][sy][1]*psi[sx][sy][0]; psi[sx][sy][0]=wr; psi[sx][sy][1]=wi; } The program qd.c simulates an electron incident on a potential barrier of height BH and width BW, see the figure below. In addition, an edge potential of height EH is applied at i = 1 or NX or j = 1 or NY in v[i][j]. The potential v[i][j] = 0 at all the other mesh points. Figure: potential The energy function. 8 In classical mechanics, a particle coming from one side of the potential barrier with a higher kinetic energy than BH will pass through the barrier to the other side; otherwise, the particle will bounce back at the barrier. In quantum mechanics, a part of the electron wave function is transmitted through the barrier and the other part is reflected at the barrier. ND: The number of spatial dimensions = 2. al[ND][2][2]: al[0|1][0|1][0|1] is the x|y-direction, half (t/2)|full (t)-step, real|imaginary-part of the diagonal element of the kinetic propagator, see Eqs. (21) and (22). bux|y[2][NX+2|Y+2][2]: bux|y[0|1][i][0|1] is the x|y-direction, half (t/2)|full (t)-step, real|imaginary-part of the upper off-diagonal kinetic propagator on mesh i, see Eqs. (21) and (22). blx|y[2][NX+2|Y+2][2]: blx|y[0|1][i][0|1] is the x|y-direction, half (t/2)|full (t)-step, real|imaginary-part of the lower off-diagonal kinetic propagator on mesh i, see Eqs. (21) and (22). The 22 block-diagonal form of the kinetic propagator, Eq. (21), can be handled (U x(half) )i, j = mod(i,2),0i1, j + 2+i, j + mod(i,2),1i +1, j 2 2 (U x(full))i, j = 1mod(i,2),1i1, j + 1+i, j + 1mod(i,2),0i +1, j (32) (33) where mod(i,2),0 = 1 (if mod(i, 2) = 0) and 0 (else), etc. The above kinetic propagator is used in qd.c to update the wave function as follows, where d (= 0 for x; 1 for y) is the direction and t (= 0 for t/2 half; 1 for tfull) is the time step: /* WRK|PSI holds the new|old wave function */ for (sx=1; sx<=NX; sx++) for (sy=1; sy<=NY; sy++) { wr=al[d][t][0]*psi[sx][sy][0]-al[d][t][1]*psi[sx][sy][1]; wi=al[d][t][0]*psi[sx][sy][1]+al[d][t][1]*psi[sx][sy][0]; if (d==0) { wr+=(blx[t][sx][0]*psi[sx-1][sy][0]-blx[t][sx][1]*psi[sx-1][sy][1]); wi+=(blx[t][sx][0]*psi[sx-1][sy][1]+blx[t][sx][1]*psi[sx-1][sy][0]); wr+=(bux[t][sx][0]*psi[sx+1][sy][0]-bux[t][sx][1]*psi[sx+1][sy][1]); wi+=(bux[t][sx][0]*psi[sx+1][sy][1]+bux[t][sx][1]*psi[sx+1][sy][0]); } else if (d==1) { wr+=(bly[t][sy][0]*psi[sx][sy-1][0]-bly[t][sy][1]*psi[sx][sy-1][1]); wi+=(bly[t][sy][0]*psi[sx][sy-1][1]+bly[t][sy][1]*psi[sx][sy-1][0]); wr+=(buy[t][sy][0]*psi[sx][sy+1][0]-buy[t][sy][1]*psi[sx][sy+1][1]); wi+=(buy[t][sy][0]*psi[sx][sy+1][1]+buy[t][sy][1]*psi[sx][sy+1][0]); } wrk[sx][sy][0]=wr; wrk[sx][sy][1]=wi; } /* Copy the new wave function back to PSI */ for (sx=1; sx<=NX; sx++) for (sy=1; sy<=NY; sy++) for (s=0; s<=1; s++) psi[sx][sy][s]=wrk[sx][sy][s]; INITIAL WAVE FUNCTION We choose the initial wave function at time t = 0 to be y (x x ) 2 0 (x, y,t = 0) = C exp exp(ik0 x ) sin . L 4 2 y To understand the meaning of this wave function, consider (34) 9 (x, y) = C exp(ik0 x ) , where C = L x Ly is the normalization constant such that |(x, y) |2 = 0 x dx 0 y dyC 2 (cos 2 (k0 x) + sin 2 (k0 x)) = 1. 0 x dx 0 y dy Then L L L L (35) (36) 1 2 2 H(x, y) = 2 + 2 C exp(ik 0 x ) 2 x y C d2 exp(ik 0 x ) 2 dx 2 ik C d = 0 exp(ik0 x ) 2 dx = , (37) 2 k0 C k2 exp(ik 0 x ) = 0 (x, y) 2 2 and thus the expectation value of its energy is = E 0 = 0 x dx 0 y dy * (x, y)H(x, y) = 0 x dx 0 y dy * (x, y) = 2 k0 2 L L 2 k0 (x, y) . 2 L L (38) or k0 = 2E 0 . We can show that the following traveling wave function is a solution to the time-dependent Schrdinger equation, if the potential energy function is 0: (x, y) = C exp(ik0 ( x v 0 t )) , where v 0 = E 0 /k0 is the wave speed. (39) The last factor in Eq. (34) also satisfies the Schrdinger equation and follows the boundary = condition, (x,0) = (x,Ly ) 0 , which is required if there is a very high potential barrier at y = 0 and Ly (the electron then cannot exist). Finally, the first Gaussian factor in Eq. (34) acts to localize the wave function around x = x0 and spread . 10 Quantum Dynamics Basics IIOne Dimensional System This lecture note explains the quantum dynamics (QD) simulation program, qd1.c, which simulates the time evolution of a wave function in one spatial dimension. Here, we only highlight differences from the two-dimensional QD simulation program, qd.c, explained in the lecture note on Quantum Dynamics Basics. Schrdinger Wave Equation WAVE EQUATION The program qd1.c simulates the time evolution of a wave function in one-dimensional space, 0 x Lx (Lx is the system size). The one-dimensional wave function, (x,t) = Re(x,t) + iIm(x,t) C (where i = 1), satisfies the normalization condition: (1) The time evolution of the wave function, subjected to a time-independent, real-valued potential, V (x) , is described (in the dimensionless form as explained in the lecture note) by the time-dependent Schrdinger equation: (2) i (x,t) = H(x,t) . t In Eq. (2), the Hamiltonian operator, H, is defined as 1 2 + V (x) . 2 x 2 = Tx + V 2 dx (x,t) = 1. H = (3) Boundary condition: We impose the periodic boundary condition on the wave function such that DISCRETIZATION (x + L x ) = (x). (4) The wave function is discretized on a regular mesh of size x. Here x = Lx/Nx, and Nx is the number of mesh points. We denote the discretized wave function as j = (jx), see the figure below. Figure: One-dimensional mesh, on which the wave function is discretized. In the discretized form, the Hamiltonian operator in Eq. (3) acts as (H ) j = (Tx ) j + (V ) j , where (5) 11 1 j1 2 j + j +1 (Tx ) j = 2 (x) 2 (V ) = V j j j (6) and the potential-energy function is discretized as Vj = V(jx). Numerical Integration of Schrdinger Equation The time evolution of the wave function is formally written down as (t + t) = exp(iHt )(t) , (7) where we omit the indices for simplicity. Here the exponential function of an operator is defined as a series expansion, 1 n exp(iHt ) = (iHt ) . (8) n=0 n! In the split-operator method, the wave function is propagated for a small time interval, t, as (t + t) = exp(iVt /2) exp(iTx t ) exp(iVt /2)(t) + O [t ] . In Eq. (9), the application of the potential propagator, exp(-iVt/2), is straightforward. ( 3 ) (9) (exp(iVt /2) ) j it (it /2) 2 2 = j V (V ) j + 2 2! ( ) j + (10) 2 it (it /2) 2 = j V j j + Vj j +. 2 2! = exp iV j t /2 j () ( ) Note that, for real number a, exp(ia) = 1+ ia + 1 1 1 1 (a 2 ) + (ia 3 ) + (a 4 ) + (ia 5 ) + 2! 3! 4! 4! 12 14 1 1 = 1 a + a + + i a a 3 + a 5 + . 2! 4! 3! 5! = cos(a) + isin(a) (11) Using Eq. (11) in Eq. (10), (exp(iVt /2) ) j = [cos(V jt /2) isin(V jt /2)][Re j + iIm j ] = cos V j t /2 Re j + sin V j t /2 Im j ) ( )] [( +i[cos(V j t /2) Im j sin(V j t /2) Re j ] . (12) To apply the kinetic propagator, exp(-iTxt), we use the space-splitting method (SSM). To understand the SSM, we first note that the operation of Tx on j is expressed as Tx j = b j1 + 2a j + b j +1 (13) where 12 a = 1/2 x 2 () b = 1/2(x ) 2 (14) The operation of Tx on j thus amounts to the multiplication of a tridiagonal matrix, 2a b b 2a b b 2a b Tx = . b 2...

Find millions of documents on Course Hero - Study Guides, Lecture Notes, Reference Materials, Practice Exams and more. Course Hero has millions of course specific materials providing students with the best way to expand their education.

Below is a small sample set of documents:

USC - CS - 596
Parallel Quantum Dynamics Aiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern California Em
USC - CS - 596
Hybrid MPI+OpenMP Parallel MD Aiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern Californi
USC - CS - 596
Optimizing Molecular Dynamics Aiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern Californi
USC - CS - 596
Advanced Topics in Parallel Molecular Dynamics Aiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of S
USC - CS - 596
Divide-&amp;-Conquer Parallelism Aiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern California
USC - CS - 596
CSCI596: Scientic Computing &amp; VisualizationSummaryAiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University o
USC - CS - 596
Least Square Fit of a LineProblem: Given a set of N pairs of numbers, {(xi, yi) | i = 1, ., N}, what is the best linear fit, y = ax + b,Nin the sense that it minimizes the square error, S = # ( axi + b &quot; yi ) .i=12Answer: S is a quadratic fu
USC - CS - 596
How to Install OpenGLWINDOWSInstall your favorite integrated development environment (IDE). This tutorial assumes that you have Microsoft Visual Studio installed on your machine. 1. Install OpenGL OpenGL software runtime is included as part of oper
USC - CS - 596
Scientific Virtual-Reality ApplicationsThis lecture note describes interactive three-dimensional visualization of scientific datasets, using the earliest immersive and interactive visualization platforms, CAVE and ImmersaDesk, and the associated CAV
USC - CS - 596
Massive Dataset VisualizationAiichiro Nakano Collaboratory for Advanced Computing &amp; Simulations Dept. of Computer Science, Dept. of Physics &amp; Astronomy, Dept. of Chemical Engineering &amp; Materials Science University of Southern California Email:
USC - CS - 596
Ashish Sharma sharmaa@usc.edu Collaboratory for Advanced Computing and Simulations Dept. of Computer Science Aiichiro Nakano anakano@usc.edu Dept. of Computer Science Rajiv K. Kalia rkalia@usc.edu Dept. of Physics &amp; Astronomy Priya Vashishta priyav@u
USC - CS - 596
A Spatially Decomposed Parallel Visualization AlgorithmInternational Journal of Computational Science 1992-6669 (Print) 1992-6677 (Online) Global Information Publisher 2007, Vol. 1, No. 4, 407-421ParaViz: A Spatially Decomposed Parallel Visualiza
USC - CS - 653
International Journal of Parallel Programming, Vol. 29, No. 3, 2001Improving Memory Hierarchy Performance for Irregular Applications Using Data and Computation Reorderings 1John Mellor-Crummey, 2, 4 David Whalley, 3 and Ken Kennedy 2Received Octo
USC - CS - 653
124IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING,VOL. 13,NO. 1, JANUARY/FEBRUARY 2001Analysis of the Clustering Properties of the Hilbert Space-Filling CurveBongki Moon, H.V. Jagadish, Christos Faloutsos, Member, IEEE, and Joel H. Salt
USC - CS - 653
Cache-Oblivious AlgorithmsEXTENDED ABSTRACT SUBMITTED FOR PUBLICATION.Matteo Frigo Charles E. Leiserson Harald Prokop Sridhar Ramachandran MIT Laboratory for Computer Science, 545 Technology Square, Cambridge, MA 021393 0 8 ( B8 &amp; 8 3 &amp; 2
USC - CS - 596
CSCI596 (Scientific Computing and Visualization) Assignment 1Molecular Dynamics Due: September 15 (Mon), 2008, at the class The purpose of this assignment is to familiarize yourself with the simple molecular dynamics (MD) program, md.c (and its linke
USC - CS - 596
CSCI596 Assignment 2Message Passing Interface Due: September 24 (Wed), 2008, at the class In this assignment, you will write your own global summation program (equivalent to using MPI_Send and MPI_Recv. Your program should run on P = 2l processors (l
USC - CS - 596
CSCI596 Assignment 3Parallel Computation of Due: October 1 (Wed), 2008 Part I: Programming Write a message passing interface (MPI) program, global_pi.c, to compute the value of based on the lecture note on Parallel Computation of Pi and using the g
USC - CS - 596
CSCI596 Assignment 4Parallel Molecular Dynamics Due: October 8 (Wed), 2008 The purpose of this assignment is to get familiar with the message-passing scheme used in the parallel molecular dynamics (MD) program, pmd.c, and asynchronous messages in the
USC - CS - 596
CSCI596 Assignment 5: Molecular Dynamics Animation Due: October 22 (Wed), 2008 Combine md.c and atomv.c to write a C/OpenGL program that animates molecular dynamics (MD) simulation in real time. (Follow the lecture note on Visualizing Molecular Dynam
USC - CS - 596
CSCI596 Assignment 6: Parallel Quantum Dynamics Due: November 3 (Mon), 2008 (at the class) Parallelize the one-dimensional quantum dynamics (QD) simulation program qd1.c, using MPI. Use spatial decomposition, so that processor p [0, P-1] (P is the n
USC - CS - 596
CSCI596 (Scientific Computing and Visualization) Assignment 7 Hybrid MPI+OpenMP Parallel Molecular Dynamics Due: November 12 (Wed), 2008 (at the class) 1. Write a hybrid MPI+OpenMP parallel molecular dynamics (MD) program (name it hmd.c), starting fr
USC - CS - 596
CSCI596 (Scientific Computing and Visualization) Final Project Anything Related to What You Have Learned in the Class Due: December 17 (Wed), 2008 Submit the following project by Wednesday, December 17. In addition, at 3:30-4:50pm on Wednesday, Decem
USC - CS - 596
Guidelines for the Final Project I. II. Programming Critical review (&gt;2-3 pages) Dont repeat what the paper says. 1. 2. 3. 4. * III. Problem: Whats the problem? Method: How to solve it? Results: Bottom line? So what? Critique: Flaw? Improvement (how
USC - CS - 653
84Computer Physics Communications 63 (1991) 8494 North-HollandVisualizing quantum scattering on the CM-2 supercomputerJohn L. Richardson1Received 2 February 1990 Thinking Machines Corporation, 245 First Street, Cambridge, MA 02142-1214, USAWe
USC - CS - 653
CSCI653: High Performance Computing &amp; SimulationsAiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of
USC - CS - 653
Molecular Dynamics I: PrinciplesBasics of the molecular-dynamics (MD) method1-3 are described, along with corresponding data structures in program, md.c.Newtons Second Law of MotionTRAJECTORY, COORDINATE, AND ACCELERATION Physical system = a set
USC - CS - 653
Minimal Complex Analysis Complex function: A mapping from a complex variable z = x + iy (i = f(z) C.&quot;1) to a complex numberDifferentiation: A complex function f(z) at z is differentiable if the quantity ! f (z + &quot;z) # f (z)&quot;zconverges to a
USC - CS - 653
JOURNALOF COMPUTATIONALPHYSICS73, 315-348 (1987)A Fast Algorithmfor ParticleANDSimulations*L. GREENCARDV. ROKHLINDepartment of Computer Science, Yale Lnipersiry, New Haven, Connecticut 06520 Received June 10. 1986; revised February
USC - CS - 653
ELSEYIER2s -/.-I@Computer Physics Communications Computer Physics Communications ( 1997)59-69 104Parallel multilevel preconditioned conjugate-gradient approach to variable-charge molecular dynamicsAiichiro Nakano Depurtment of Computer Scienc
USC - CS - 653
Computer Physics Communications 153 (2003) 445461 www.elsevier.com/locate/cpcScalable and portable implementation of the fast multipole method on parallel computers Shuji Ogata a , Timothy J. Campbell b , Rajiv K. Kalia c,d , Aiichiro Nakano c,d,
USC - CS - 653
Reversiblemultipletime scale moleculardynamicsM. Tuckermar?) G. J. Martynaand B. J. BerneDepartment of Chemistry, Columbia University, New York, New York 10027 Department of Chemistry, University of Pennsylvania, Philadelphia, Pennsylvani
USC - CS - 653
Computer Physics CommunicationsELSEVIERComputer Physics Communications 83 (1994) 197214Multiresolution molecular dynamics algorithm for realistic materials modeling on parallel computersAiichiro Nakano, Rajiv K. Kalia, Priya VashishtaConcurrent
USC - CS - 653
Message Passing Interface (MPI) ProgrammingMPI (Message Passing Interface) is a standard message passing system that enables us to write and run applications on parallel computers. In 1992, MPI Forum was formed to develop a portable message passing
USC - CS - 653
OpenMP ProgrammingAiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern California Email: ana
USC - CS - 653
Hybrid MPI+OpenMP Parallel MDAiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern California
USC - CS - 653
Parallel Molecular DynamicsThis chapter explains the example parallel MD program, pmd.c, in detail.Spatial Decomposition Spatial decomposition: The physical system to be simulated is partitioned into subsystems of equal volume. Processors in a pa
USC - CS - 653
Parallel Molecular DynamicsAiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern California E
USC - CS - 653
Scalability Metrics for Parallel Molecular DynamicsParallel EfficiencyWe define the efficiency of a parallel program running on P processors to solve a problem of size W. Let T(W, P) be the execution time of this parallel program. Speed of the prog
USC - CS - 653
Anton, a Special-Purpose Machine for Molecular Dynamics SimulationDavid E. Shaw, Martin M. Deneroff, Ron O. Dror, Jeffrey S. Kuskin, Richard H. Larson, John K. Salmon, Cliff Young, Brannon Batson, Kevin J. Bowers, Jack C. Chao, Michael P. Eastwood,
USC - CS - 653
A Fast, Scalable Method for the Parallel Evaluation of Distance-Limited Pairwise Particle InteractionsDAVID E. SHAWD. E. Shaw Research and Development, LLC and Center for Computational Biology and Bioinformatics, Columbia University, 120 W. 45th S
USC - CS - 653
Divide-and-Conquer Parallelization ParadigmDivide-and-Conquer Simulation Algorithms Divide-and-conquer (DC) algorithms: Recursively partition a problem into subprogram of roughly equal size. If subprogram can be solved independently, there is a pos
USC - CS - 653
Divide-&amp;-Conquer ParallelismAiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern California
USC - CS - 653
Computational Materials Science 38 (2007) 642652 www.elsevier.com/locate/commatsciA divide-and-conquer/cellular-decomposition framework for million-to-billion atom simulations of chemical reactionsAiichiro Nakano a,*, Rajiv K. Kalia a, Ken-ichi No
USC - CS - 653
DE NOVO ULTRASCALE ATOMISTIC SIMULATIONS ON HIGH-END PARALLEL SUPERCOMPUTERSAiichiro Nakano Rajiv K. Kalia1 Ken-ichi Nomura1 Ashish Sharma1, 2 Priya Vashishta1 Fuyuki Shimojo1,3 4 Adri C. T. van Duin 4 William A. Goddard, III 5 Rupak Biswas Deepak S
USC - CS - 653
Load BalancingAiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern California Email: anakano
USC - CS - 653
Lanczos Method for EigensystemsAiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern Californ
USC - CS - 653
Supplementary Derivations for the Lanczos-Algorithm LectureSpectral representation The eigenvalues and eigenvectors satisfyn n\$ Aij q (&quot; ) = #&quot; qi(&quot; ) = \$ qi(&quot; ) #%&amp;%&quot; , jj=1%=1()(1)where &quot;#\$ = 1 (\$ = #); 0 (\$ % #). Define an orthogona
USC - CS - 653
C3P 913June 1990Performance of Dynamic Load Balancing Algorithms for Unstructured Mesh CalculationsRoy D. WilliamsConcurrent Supercomputing Facility California Institute of Technology Pasadena, CaliforniaAbstract If a nite element mesh has a
USC - CS - 653
SIAM J. SCI. COMPUT. Vol. 20, No. 1, pp. 359392c 1998 Society for Industrial and Applied MathematicsA FAST AND HIGH QUALITY MULTILEVEL SCHEME FOR PARTITIONING IRREGULAR GRAPHSGEORGE KARYPIS AND VIPIN KUMAR Abstract. Recently, a number of researc
USC - CS - 653
Applied Numerical Mathematics 52 (2005) 133152www.elsevier.com/locate/apnumNew challenges in dynamic load balancingKaren D. Devine a, , Erik G. Boman a , Robert T. Heaphy a , Bruce A. Hendrickson a , James D. Teresco b,1 , Jamal Faik c , Joseph E
USC - CS - 653
Hypergraph-based Dynamic Load Balancing for Adaptive Scientic ComputationsUmit V. Catalyurek, Erik G. Boman, Karen D. Devine, Doruk Bozda , Robert Heaphy, g and Lee Ann Riesen Ohio State University Sandia National Laboratories Dept. of Biomedical
USC - CS - 653
Optimizing Molecular DynamicsAiichiro NakanoCollaboratory for Advanced Computing &amp; Simulations Department of Computer Science Department of Physics &amp; Astronomy Department of Chemical Engineering &amp; Materials Science University of Southern California
USC - CS - 653
Improving Memory Hierarchy Performance for Irregular Applications*John Mellor-Crummeyt, T Department of Computer Science, MS 132 Rice University 6100 Main Houston, TX 77005 David Whalleyz, Ken Kennedy?Cjohnmc,ken}@cs.rice.edu AbstractThe gap betw
USC - CS - 653
Metrics and Models for Reordering TransformationsMichelle Mills StroutMathematics and Computer Science Division Argonne National Laboratory Argonne, IL 60439 USAPaul D. HovlandMathematics and Computer Science Division Argonne National Laboratory
USC - CS - 653
SIAM REVIEW Vol. 46, No. 1, pp. 345c 2004 Society for Industrial and Applied MathematicsRecursive Blocked Algorithms and Hybrid Data Structures for Dense Matrix Library SoftwareErik Elmroth Fred Gustavson Isak Jonsson Bo K gstrom aAbstract. Ma
USC - CS - 653
Lattice-Boltzmann (LB) Fluid Simulation on a Playstation3 (PS3) ClusterKen-ichi Nomura, Liu Peng &amp; Richard SeymourCollaboratory for Advanced Computing &amp; Simulations Dept. of Computer Science, Dept. of Physics &amp; Astronomy, Dept. of Chemical Engineer
USC - CS - 653
SCOP3A Rough Guide to Scientic Computing On the PlayStation 3Technical Report UT-CS-07-595 Version 0.1by Alfredo Buttari Piotr Luszczek Jakub Kurzak Jack Dongarra George Bosilca Innovative Computing Laboratory University of Tennessee Knoxville Ap
USC - CS - 653
Parallel Lattice Boltzmann Flow Simulation on a Low-cost PlayStation3 ClusterInternational Journal of Computational Science 1992-6669 (Print) 1992-6677 (Online) www.gip.hk/ijcs 2008 Global Information Publisher (H.K) Co., Ltd. All rights reserved.