Propagating the Time Dependent Schroedinger Equation B. I. Schneider Division of Advanced Cyberinfrastructure National Science Foundation H (r ) T V( r ) National Science Foundation September 6, 2013 What Motivates Our Interest •Novel light sources: ultrashort, intense pulses Nonlinear (multiphoton) laser-matter interaction Attosecond pulses Free electron lasers (FELs) probe and control electron Extreme intensities dynamics XUV + IR pump-probe Multiple XUV photons Basic Equation i ( r , t ) H ( r , t ) (r , t ) 0 t Where 2 H (r , t ) 2 2 i m i i Possibly Non-Local or Non-Linear V( r, t) Properties of Classical Orthogonal Functions Orthonorma lity w.r.t . some positive weight function. b χ χ dx w(x) χ (x) χ (x) n m n m n, m a The functions satisfy a three term recursion relationsh ip of the form; β χ (x) (x α )χ (x) β χ (x) n n n 1 n 1 n 1 n 2 The recursion coefficien ts may be computed using the Lanczos procedure A set of Gauss quadrature points, x and weights, w may be found which i i exactly integrate any polynomial integrand of order (2n - 1) or less with respect to the weight function. The points and weights may be found by diagonaliz ing the tridiagon al matrix made up of the and coefficien ts. Completene ss n (x ) n (x) (x - x) ; n n ( x ) w1/2 (x) χ n (x) More Properties A discrete orthonorma lity relationsh ip Note that the orthonorma lity integral can be performed exactly by p - point Gauss quadrature s for all where q p q p n m wi χ n (x i )χ m (x i ) δn, m i 1 This is true because the integrand is a polynomial which can be integrated exactly by the quadrature . Corollary Given an expansion, p 1/2 Ψ(x) w (x) (x) c (x) q q q 1 b p c dx w(x) χ (x) (x) χ (x ) w (x ) q q q i i i i 1 a Matrix Elements Consider, a matrix element of the potential, Vq,q' q V(x) q' Conceptual ly, this matrix element may be evaluated if we know the matrix representa tion of the position operator. Then, V V (x) as long as the basis set is complete. This remains quite useful even for finite basis sets and suggests that an excellent approximat ion to the matrix is obtained, Vq,q' Tq,i V(x i ) Tq' ,i i where T is the transform ation from the original representa tion to one which diagonaliz es x. Note, that this looks like a quadrature formula. Properties of Discrete Variable Representation Define a new set of " coordinate " functions, p u i (x) w i q (x) q (x i ) q 1 with the properties that, u i (x j ) i, j wi b ; dx w(x) u i (x)u j (x) i, j a and u i x u j x i i, j Consider t he matrix element Fi, j u i F(x) u j In general this will not be equal to Fi, j F(x i ) i, j unless the basis/or quadrature is complete. (Think power series expansion and matrix mutiply) In the DVR, its is ASSUMED that this is true. In practice it appears to be an excellent approximat ion. Its Actually Trivial A simple representa tion (x-x ) 1 j u i (x) W j i ( xi - x j ) i where x are the Gauss quadrature points. i Matrix elements of the derivative operators may be evaluated using the quadrature rule, but its trivial . For Cartesian coordinates : d2 d2 ui u j w iui ( x i ) u j 2 2 dx dx x xi Multidimensional Problems Tensor Product Basis Consequences i, j, k H l, m, n i T l j,m k,n j T m i,l k,n k T n i,l j,m i, j, k V l, m, n i,l j,m k,n Multidimensional Problems out Vi,j,k = å i,j,k H l,m,n in Vl,m,n = l,m,n å iTl in Vl,j,k l +å j T m in Vi,m,k m +å k T n in Vi,j,n n in + i,j,k V i,j,k Vi,j,k Nested sums. Two Electron matrix elements also ‘diagonal” using Poisson equation Finite Element Discrete Variable Representation • Properties •Space Divided into Elements – Arbitrary size •“Low-Order” Lobatto DVR used in each element: first and last DVR point shared by adjoining elements F ( x) i n i 1 1 ( f ( x) f i n i 1 1 w w i n ( x) ) • • Sparse Representations – N Scaling Close to Spectral Accuracy •Elements joined at boundary – Functions continuous but not derivatives •Matrix elements requires NO Quadrature – Constructed from renormalized, single element, matrix elements Finite Element Discrete Variable Representation • Structure of Matrix h11 h21 h12 h22 h13 h23 h14 h24 h31 h41 h32 h42 h33 h43 h34 h44 h45 h46 h54 h64 h55 h65 h56 h66 h67 h76 h77 Time Propagation Method H(t)t (r, t + t ) = exp(-i ) (r, t ) Diagonalize Hamiltonian in Krylov basis •Few recursions needed for short time- Typically 10 to 20 via adaptive time stepping •Unconditionally stable • Major step - matrix vector multiply, a few scalar products and diagonalization of tri-diagonal matrix Putting it together for the He Code NR1 NR2 Angular Linear scaling with number of CPUs Limiting factor: Memory bandwidth XSEDE Lonestar and VSC Cluster have identical Westmere processors Comparison of He Theoretical and Available Experimental Results NSDI -Total X-Sect Considerable discrepancies! Rise at sequential threshold Extensive convergence tests: angular momenta, radial grid, pulse duration (up to 20 fs), time after pulse (propagate electrons to asymptotic region) error below 1% R (a.u.) R (a.u.) FIG. 4. Left : Ionizat ion probabilit ies (left scale) for H +2 in laser pulses at phot on energies of 1.50 and 1.18 a.u. for t ime durat ions of 10 and 5 cycles, respect ively . T he molecular axis is chosen along t he linear laser polarizat ion vect or. T he peak int ensity is 1014 W / cm2 . T he elect ronic energy (dot -dot -dashed line) versus t he int ernuclear dist ance R is shown on t he right scale. Right : Transit ion moment s (left scale) in t he parallel geomet ry from t he channels ` = 1, 3, and 5 at t he phot on energy of 1.50 a.u. T he t ot al squared cont ribut ion from all t hree channels is shown on t he right scale (solid curve). T he vert ical dashed line indicat es t he R value where M 1, 0 vanishes. E 2 (eV) 20 10 15 10 5 5 0 0 0 5 10 E 1 (eV) 15 2 ✓1 = 30◦ TDCS (10− 55cm4 s/ sr2 eV) 25 15 TDCS (10− 55cm4 s/ sr2 eV) Two-Photon Double Ionization in FE-DVR ECS (/ 2) TDCC (⇥2) 1 0 0 60 120 180 240 ✓2 (deg) 300 360 40 ✓1 = 30◦ 30 20 10 0 0 60 120 180 240 ✓2 (deg) 300 360 FIG. 5. Left : Energy probability dist ribut ion of two eject ed elect rons when t he laser polarizat ion axis is perpendicular t o t he molecular axis for a sine-squared laser pulse of 10 opt ical cycles, a cent ral energy of 30 eV , and a peak int ensity of 1014 W / cm2 . T he color bars correspond t o mult iples of 10− 7 eV − 2 . A lso shown are t he coplanar T DCS for two-phot on double ionizat ion of H 2 at equal-energy sharing (E 1 = E 2 = 4.3 eV ) of t he two eject ed elect rons in t he parallel (cent er) and again t he perpendicular geomet ry (right ) for a det ect ion of one elect ron being 30◦ relat ive t o t he laser polarizat ion. T he T DCS result s of Colgan et al. [69] and M orales et al. [70] for t he parallel geomet ry (cent ral panel) were mult iplied by t he scaling fact ors indicat ed in t he legend. our resultspectral s and t hose from ot her recent calculat ions by Colgan et al.of [69] the and Morales et al. [70]. can Our analysis [8] The Characteristics Pulse suggest s t hat t he reason may not be t he numerics (albeit t he st rong emission of t he second elect ron seen in [69] in t he vicinity of t he eject ion angle of t he first Critical one seems quest ionable), but inst ead a consequence of t he t imebe dependent t reat ment wit h a relat ively broad spect rum of phot on energies vs. t he e↵ect ively t ime-independent Can We Do Better ? How to efficiently approximate the integral is the key issue