Maxeler@FFH: Selected FFH Applications 1/12 MAXELER @ FFH ŠĆEPAN MILJANIĆ, EPAN@FFH.BG.AC.RS MILENA PETKOVIĆ, MILENA@FFH.BG.AC.RS ALEKSANDAR JOVIĆ, ALEKSANDAR.JOVIC@FFH.BG.AC.RS 2/12 A Short Survey of FFH Algorithms Suitable for DataFlow Technology Quantum Chemistry Chemical Kinetics Analysis of Fluids Statistical Thermodynamics Spectroscopy 3/12 Condition for an Algorithm to be Suitable for DataFlow The FFH Viewpoint: a. BigData b. Over 95 % of run time in loops c. Reusability of the data d. n/a e. n/a f. n/a 4/12 Quantum Chemistry Time-independent Schrödinger equation: Ĥ E Goal: optimization of the structure! Method of steepest descent – the next point xn+1 is chosen in the direction of the negative gradient at xn: xn1 xn sn E ( xn ) Conjugate gradient methods – gradients at the last two points are taken into account: xn1 f xn , E ( xn ), E ( xn1 ) Newton-Raphson method – in addition to the gradient, the Hessian is computed at each point: 1 xn 1 xn H ( xn ) E ( xn ) 5/12 Chemical Kinetics System of partial differential equations: used for finding numerical solutions to sets of coupled chemical reactions. Euler method yn 1 yn hf ( xn , yn ) Runge-Kutta algorithm k1 hf ( xn , yn ) k2 hf ( xn 12 h, yn 12 k1 ) yn1 yn k2 O(h ) 3 Taylor theorem dN / dt a bN cN 2 dN 3 ... 6/12 Analysis of Fluids Monte-Carlo methods n b-a I g ( xi ) n i 1 b I g ( x)dx a Korteweg–de Vries equation (KdV equation) 2 P ( , , 2n1 x x ,...)dx dPn 1 n 2 Pn Pi Pn 1i dx i 1 Vlasov equation for classical fluids r 1 ( p, r; t ) mP r 1 ( p, r; t ) 7/12 r 1 z (r r ) 1 (r ; t )dr P 0 ( p) Statistical Thermodynamics System of a large number of statistical methods Euler-Lagrange equations d L dt qi L qi Boltzmann H-function H B ( p1 , p2 ,...) pk ln pk k Schmidt recurrence relation n p ( N )Z N e 8/12 e p Z N a 1 n p ( N 1) ZN Spectroscopy Fourier transform f ( x) ˆf ( )e 2ix d Taylor Series Expansion Bz ( x, y , z ) i , j , k x y z i j k i , j ,k Green's function b (G f )( x) G ( x, y) f ( y)d ( y) a 9/12 Synergy Possibilities: FFH and ETF HW @ ETF HW @ FFH Periodical Joint Meetings ETF speed up of concrete FFH calculations 10/12 Why all of this? 11/12 http://www.tcm.phy.cam.ac.uk/~mdt26/qmc_projects/dft/functionals.html ŠĆEPAN MILJANIĆ, EPAN@FFH.BG.AC.RS MILENA PETKOVIĆ, MILENA@FFH.BG.AC.RS ALEKSANDAR JOVIĆ, ALEKSANDAR.JOVIC@FFH.BG.AC.RS 12/12