1 / 44

Computer Science Colloquium

Computer Science Colloquium. Large Scale Computational Challenges in Materials Science *. * Work supported by NSF under grants NSF/ITR-0082094, NSF/ACI-0305120 and by the Minnesota Supercomputing Institute. C. Bekas: CS Colloquium. Introduction and Motivation.

thilson
Download Presentation

Computer Science Colloquium

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Computer Science Colloquium Large Scale Computational Challenges in Materials Science* *Work supported by NSF under grants NSF/ITR-0082094, NSF/ACI-0305120 and by the Minnesota Supercomputing Institute

  2. C. Bekas: CS Colloquium Introduction and Motivation • Computational Materials Science Target Problem • …predict the properties of materials… • How? • AB INITIO calculations:Simulate the behavior of materials at the atomic level, by applying the basic laws of physics (Quantum mechanics) • What do we (hope to) achieve? • Explain the experimentally found properties of materials • Engineer new materials with desired properties Applications:…numerous (some include) • Semiconductors, synthetic light weight materials • Drug discovery, protein structure prediction • Energy: alternative fuels (nanotubes, etc) Large Scale Computational Challenges in Materials Science

  3. C. Bekas: CS Colloquium In this talk • Introduction to the mathematical formulation of ab initio calculations… • in particular…the Density Functional Theory (DFT)…formulation • Identify the computationally intensive “spots”… Large scale eigenvalue problems are central… • symmetric/hermitian problems • very large number of eigenvalues/vectors required…so • reorthogonalization and synchronization costs dominate… • limiting the feasible size of molecules under study Alternative…Automated Multilevel Substructuring (AMLS) • Significantly limits reorthogonalization costs… • can attack very large problems…when O(1000) eigs. are required… Techniques that avoid the eigenvalue problem altogether… • Monte-Carlo simulations for the diagonal of a matrix (charge densities) Large Scale Computational Challenges in Materials Science

  4. C. Bekas: CS Colloquium Mathematical Modelling: Wave Function We seek to find the steady state of the electron distribution • Each electron eiis described by a corresponding wave function ψi … • ψiis a function of space (r)…in particular it is determined by • The position rk of all particles (including nuclei and electrons) • It is normalized in such a way that • Max Bohr’s probabilistic interpretation: Considering a region D, then …describes the probability of electron ei being in region D. Thus: the distribution of electrons ei in space is defined by the wave function ψi Large Scale Computational Challenges in Materials Science

  5. C. Bekas: CS Colloquium Mathematical Modelling: Hamiltonian • Steady state of the electron distribution: • is it such that it minimizes the total energy of the molecular system…(energy due to dynamic interaction of all the particles involved because of the forces that act upon them) HamiltonianH of the molecular system: • Operator that governs the interaction of the involved particles… • Considering all forces between nuclei and electrons we have… Hnucl Kinetic energy of the nuclei He Kinetic energy of electrons UnuclInteraction energy of nuclei (Coulombic repulsion) VextNuclei electrostatic potential with which electrons interact UeeElectrostatic repulsion between electrons Large Scale Computational Challenges in Materials Science

  6. C. Bekas: CS Colloquium Mathematical Modelling: Schrödinger's Equation Let the columns of Ψ: hold the wave functions corresponding the electrons…Then it holds that • This is an eigenvalue problem…that becomes a usual… • “algebraic” eigenvalue problem when we discretize i w.r.t. space (r) • Extremely complex and nonlinear problem…since • Hamiltonian and wave functions depend upon all particles… • We can very rarely (only for trivial cases) solve it exactly… Variational Principle (in simple terms!) Minimal energy and the corresponding electron distribution amounts to calculating the smallest eigenvalue/eigenvector of the Schrödinger equation Large Scale Computational Challenges in Materials Science

  7. C. Bekas: CS Colloquium Schrödinger's Equation: Basic Approximations Multiple interactions of all particles…result to extremely complex Hamiltonian…which typically becomes huge when we discretize Thus…a number of reasonable approximations/simplifications have been considered…with negligible effects on the accuracy of the modeling: • Born-Oppenheimer: Separate the movement of nuclei and electrons…the latter depends on the positions of the nuclei in a parametric way…(essentially neglect the kinetic energy of the nuclei) • Pseudopotential approximation: Nucleus and surrounding core electrons are treated as one entity • Local Density Approximation: If electron density does not change rapidly w.r.t. sparse (r)…then electrostatic repulsion Uee is approximated by assuming that density is locally uniform Large Scale Computational Challenges in Materials Science

  8. C. Bekas: CS Colloquium Density Functional Theory High complexity is mainly due to the many-electron formulation of ab initio calculations…is there a way to come up with an one-electron formulation? Key Theory DFT: Density Functional Theory (Hohenberg-Kohn, ‘64) • The total ground energy of a system of electrons is a functional of the electronic density…(number of electrons in a cubic unit) • The energy of a system of electrons is at a minimum if it is an exact density of the ground state! • This is an existence theorem…the density functional always exists • …but the theorem does not prescribe a way to compute it… • This energy functional is highly complicated… • Thus approximations are considered…concerning: • Kinetic energy and • Exchange-Correlation energies of the system of electrons Large Scale Computational Challenges in Materials Science

  9. Kinetic energy of electronei Total potential that acts on ei at position r Energy of the i-th state of the system One electron wave function Charge density at position r C. Bekas: CS Colloquium Density Functional Theory: Formulation (1/2) Equivalent eigenproblem: Large Scale Computational Challenges in Materials Science

  10. C. Bekas: CS Colloquium Density Functional Theory: Formulation (2/2) Furthermore: Potential due to nuclei and core electrons Coulomb potential form valence electrons Exchange-Correlation potential…also a function of the charge density  Non-linearity:The new Hamiltonian depends upon the charge density  while  itself depends upon the wave functions (eigenvectors) i Thus: some short of iteration is required until convergence is achieved! Large Scale Computational Challenges in Materials Science

  11. C. Bekas: CS Colloquium Self Consistent Iteration Large Scale Computational Challenges in Materials Science

  12. C. Bekas: CS Colloquium S.C.I: Computational Considerations Large Scale Computational Challenges in Materials Science

  13. C. Bekas: CS Colloquium S.C.I: Computational Considerations Conventional approach: • Solve the eigenvalue problem (1)…and compute the charge densities… • This is a tough problem…many of the smallest eigenvalues…deep into the spectrum are required! Thus… • efficient eigensolvers have a significant impact on electronic structure calculations! Alternative approach: • The eigenvectors i are required only to compute k(r) • Can we instead approximate charge densities without eigenvectors…? • Yes…! Large Scale Computational Challenges in Materials Science

  14. C. Bekas: CS Colloquium Computational Considerations in Applying Eigensolvers for Electronic Structure Calculations Large Scale Computational Challenges in Materials Science

  15. C. Bekas: CS Colloquium The Eigenproblem Hamiltonian Characteristics • Discretization: High-order finite difference scheme…leads to • Large Hamiltonians!…typically N>100K…with significant… • number of nonzero elements (NNZ)>5M… • Hamiltonian is Symmetric/Hermitian…thus the eigenvalues are real numbers…some smallest and some larger than zero. Eigenproblem Characteristics(why it is a tough case?) • We need the algebraically smallest (leftmost) eigenvalues (and vectors) • How many? Typically a large number of them. Depending upon the molecular system under study: • for standard spin-less calculations  SixHy: (4x + y)/2 • i.e. for the small molecule Si34H36 we need the 86 smallest eigenvalues… • For large molecules, x,y>500 (or for exotic entities…quantum dots) thousands of the smallest eigenvalues are required…. • Using current state of the-art-methods we need thousands of CPU hours on DOE supercomputers…and we have to do that many times! Large Scale Computational Challenges in Materials Science

  16. C. Bekas: CS Colloquium Methods for Eigenvalues (basics only!) Eigenvalue Approximation from a Subspace Consider the standard eigenvalue problem: and let V be a thin N x k (N>>k) matrix…then approximate the original problem with: • Observe that: • Selecting V to have orthogonal columns …VTV = I … but it is expensive to come up with an orthogonal V • Set H = VTAV… then H is k x k …much smaller than N x N Large Scale Computational Challenges in Materials Science

  17. C. Bekas: CS Colloquium Subspace Methods H is much smaller than A…use LAPACK for H • How to compute orthogonal V ? • For a non-symmetric matrix A…use Gramm-Schmidt (Arnoldi)… • For a symmetric matrix A (our case) use Lanczos… • Other approaches available (i.e. Jacobi-Davidson) but still some sort of Gramm-Schmidt is required… REMINDER We need many eigenvalues/vectors O(1000)…thus V may not be that thin!!! Large Scale Computational Challenges in Materials Science

  18. NO SYNC. NO SYNC. C. Bekas: CS Colloquium Symmetric Problems: Lanczos Basic property Theoretically…(assuming no round-off errors)…Lanczos can build a very large orthogonal basis V requiring in memory only 3 columns of V at each step! • Lanczos • 1. Input: Matrix A, unit norm starting vector v0, 0 = 0, # k • 2. For j = 1,2,…,k Do • 3. wj = Avj MATRIX VECTOR • 4. wj = wj - j vj-1 DAXPY • 5. j = (wj, vj) DOT PRODUCT • 6. wj = wj - j vj DAXPY • 7. j+1 = ||wj||2. DOT PRODUCT • 8. If j+1 = 0 then STOP • 9. vj+1 = wj / j+1 DSCAL • 10. EndDO SYNC. -BCAST NO SYNC. SYNC. - BCAST NO SYNC. Large Scale Computational Challenges in Materials Science

  19. C. Bekas: CS Colloquium Lanczos in Finite Arithmetic… Round-off errors • Lanczos vectors vi quickly loose orthogonality…so that • VTV is no longer orthogonal…thus • We need to check if vj is  to previous vectors 0,1,…,j-1 • If NOT … reorthogonalize it against previous vectors (Gramm-Schmidt) • Lanczos • 1. Input: Matrix A, unit norm starting vector v0, 0 = 0, # k • 2. For j = 1,2,…,k Do • 3. wj = Avj • 4. wj = wj - j vj-1 • 5. j = (wj, vj) • 6. wj = wj - j vj • 7. j+1 = ||wj||2. • 8. If j+1 = 0 then STOP • 9. vj+1 = wj / j+1 • 10. EndDO ORTHOGONALITY IS LOST HERE…SO THESE STEPS ARE REPEATED AGAINST ALL PREVIOUS VECTORS… SELECTIVE REORTH. IS ALSO POSIBLE (SIMON, LARSEN) Large Scale Computational Challenges in Materials Science

  20. C. Bekas: CS Colloquium Practical Eigensolvers and Limitations ARPACK (Sorensen-Lehoucq-Yang): Restarted Lanczos • Remember that O(1000) eigenvalues/vectors are required…thus • we need a very long basis V…k = twice the number of eigenvalues which • will result in a large number of reorthogonalizations… • Synchronization costs – Reorthogonalization costs – and memory costs become intractable for large problems of interest… Shift-Invert Lanczos (Grimes et all) – Rational Krylov (Ruhe) • work with matrix (A-i I)-1 instead… • compute some of the eigenvalues close to i each time…thus a smaller basis is required each time…BUT • many shifts i are required… • cost of working with the different “inverses” (A-i I)-1 becomes prohibitive for (practically) large Hamiltonians… We need alternative methods that can build large projection bases without the reorthogonalization-synchronization costs Large Scale Computational Challenges in Materials Science

  21. C. Bekas: CS Colloquium Automated Multilevel Substructuring Component Mode Synthesis (CMS) (Hurty ’60, Graig-Bampton ’68) Well known alternative to Lanczos type methods. Used for many years in Structural Engineering. But it too suffers from limitations due to problem size… AMLS, (Bennighof, Lehoucq, Kaplan and collaborators)  Multilevel CMS method (solves the dimensionality problem)  Automatic computation of substructures (easy application)  Approximation: Truncated Congruence Transformation  Builds very large projection basis withoutreorthogonalization  Successful in computing thousands of eigenvalues in vibro-acoustic analysis (N>107) in a few hours on workstations (Kropp–Heiserer, 02) Spectral Schur Complements (Bekas, Saad) • Significantly improves AMLS accuracy…suitable for electronic structure calculations • (unlike AMLS) …framework for the iterative refinement of the approximations Large Scale Computational Challenges in Materials Science

  22. 2 1 Subdivide  into 2 subdomains: 1 and 2 C. Bekas: CS Colloquium Component Mode Synthesis: a model problem Consider the model problem: Y on the unit square . We wish to compute smallest eigenvalues. • Component Mode Synthesis • Solve problem on each i • “Combine” partial solutions X Large Scale Computational Challenges in Materials Science

  23. CMS: Approximation procedure. Ignore coupling! C. Bekas: CS Colloquium Component Mode Synthesis: Approximation • CMS: Approximation procedure (2) • Solve B v =  v • Remember that B is block diagonal • Thus: we have to solve smaller decoupled eigenproblems • Then, CMS approximates the coupling among the subdomains by the application of a carefully selected operator on the interface unknowns Large Scale Computational Challenges in Materials Science

  24. Block Gaussian eliminator matrix: such that: Where S = C – E> B-1 E is the Schur Complement Equivalent Generalized Eigenproblem : UTAUu= UTU u C. Bekas: CS Colloquium Recent CMS method: AMLS Large Scale Computational Challenges in Materials Science

  25. Solve decoupled problems Form basis FINAL APPROXIMATION: C. Bekas: CS Colloquium AMLS Approximation AMLS: Approximation procedure. Ignore coupling! Large Scale Computational Challenges in Materials Science

  26. B1 E1 S3 B2 S2 E1* S1 C. Bekas: CS Colloquium AMLS: Multilevel application Scheme applied recursively. Resulting to thousands of subdomains. Successful in computing thousands smallest eigenvalues in vibro-acoustic analysis with problem size N>107 (Kropp – Heiserer BMW) Large Scale Computational Challenges in Materials Science

  27. C. Bekas: CS Colloquium AMLS: Example Example: Container ship, 35K degrees of freedom (Research group of prof. H. Voss, T. U. Hamburg, Germany) Large Scale Computational Challenges in Materials Science

  28. C. Bekas: CS Colloquium AMLS: Example Example: Container ship, 35K degrees of freedom (Research group of prof. H. Voss, T. U. Hamburg, Germany) Large Scale Computational Challenges in Materials Science

  29. C. Bekas: CS Colloquium AMLS: Example • AMLS: Substructure tree (Kropp-Heiserer, BMW) • Multilevel parallelism… • Both Top-Down and Bottom-Up implementations are possible… • At each node we need to solve a linear system… • Multilevel solution of linear systems…level k depends-benefits from level k+1 Large Scale Computational Challenges in Materials Science

  30. C. Bekas: CS Colloquium Problem Set…AMLS v.s. Standard Methods Application Domains, Kropp-Heiserer, 02 NUMBER OF EIGENVALUES DEGREES OF FREEDOM Large Scale Computational Challenges in Materials Science

  31. C. Bekas: CS Colloquium AVOIDING EIGENVALUE COMPUTATIONS! Large Scale Computational Challenges in Materials Science

  32. C. Bekas: CS Colloquium Motivation • Target Problem • Estimate the Diagonal of Matrix: • The matrix is known ONLY in some “functional form”: • We have a routine MV(v,x) such that x = Av • Application in Materials Science • Estimate the diagonal of the Density matrix, which holds the Charge Densities (for which we solve the expensive eigenproblem!) • Avoid diagonalization: approximate the Density matrix by expanding the Fermi operator in a basis of Tchebychev polynomials, Jay et al (99) • Basic property to exploit: Matrix-Vector products with the approximated Density matrix are cheap to compute! Large Scale Computational Challenges in Materials Science

  33. C. Bekas: CS Colloquium Estimating the Diagonal: The direct approach • Multiply the target matrix A by the vector ei to get the i-th column of A • Inner product of the i-th column with vector ei to get the i-th element i i 0 0 . . . 1 . . 0 0 0 . . . 1 . . 0 i i Straightforward! Even embarrassingly Parallel! But: HIGH COST…WE NEED N-Matrix vector products so cost is still O(N3) Large Scale Computational Challenges in Materials Science

  34. C. Bekas: CS Colloquium A Stochastic Estimator for the Trace • Observe: the same technique can be used to estimate the trace of A, but still the cost will be O(N3) • Can we do better? Meaning: • Define special vectors vi such that only a small number m<<N will be required to obtain a good approximation of the diagonal (or the trace) • Hutchinson (90): Unbiased Stochastic Estimator of the trace for Laplacian Smoothing • Extended ideas of Girard (87) • Monte Carlo simulations with the discrete variable that assumes the values -1, 1 with probability ½ …meaning: • Define random vectors vi, i=1,…,s with entries ±1 and Large Scale Computational Challenges in Materials Science

  35. C. Bekas: CS Colloquium The Stochastic Diagonal Estimator Let the vectors vk be as before (with ±1 entries)…however they can also have other entries… Estimate the diagonal : : Component-wise multiplication : Component-wise division Large Scale Computational Challenges in Materials Science

  36. C. Bekas: CS Colloquium Estimating The Fermi Operator P: Density matrix, H: Hamiltonian Then: P = f(H), f is the Fermi-Dirac distribution: kB: Boltzmann constant, T: temperature, : Chemical potential When T0, f is treated as the Heaviside function occ 1 max Large Scale Computational Challenges in Materials Science

  37. C. Bekas: CS Colloquium Tchebychev Expansion of the Heaviside Func. The Hamiltonian is shifted and scaled so that it has eigenvalues in the interval [-1, 1] Then, we have the following approximation in the basis of Tchebychev polynomials with Jackson dampening in order to avoid Gibbs oscillations But…Tchebychev polynomials are defined recursively Large Scale Computational Challenges in Materials Science

  38. C. Bekas: CS Colloquium Tchebychev Expansion of the Heaviside Func. Due to the recursive nature of Tchebychev polynomials Matrix-Vector products with the approximation to the Density matrix is inexpensive! Thus, the Tchebyshev expansion is ideal for our stochastic and Hadamard estimators Cost of each approximate Matvec with the Density matrix: M Matrix-Vector products with the Hamiltonian and M xAXPY vector operations (O(N)) Large Scale Computational Challenges in Materials Science

  39. C. Bekas: CS Colloquium Some Experiments(2/2) Matrix: AF23650. Using only 10 random vectors in the stochastic estimator Abs. Relative error for each entry in the diagonal Large Scale Computational Challenges in Materials Science

  40. C. Bekas: CS Colloquium Some Experiments(2/2) Matrix: AF23650. Using 500 random vectors in the stochastic estimator Abs. Relative error for each entry in the diagonal Large Scale Computational Challenges in Materials Science

  41. C. Bekas: CS Colloquium Some Experiments: Density Matrix Large Scale Computational Challenges in Materials Science

  42. C. Bekas: CS Colloquium Some Experiments: SiH4 Large Scale Computational Challenges in Materials Science

  43. C. Bekas: CS Colloquium Implementation Issues – Trilinos • ab initio calculations:…many ingredients required for successful techniques • Mesh generation…discretization • Visualization of input data…results…geometry • Efficient data structures-communicators for parallel computations • Efficient (parallel) Matrix-Vector and inner products • Linear system solvers • State-of-the-art eigensolvers… A unifying software development environment will prove to be very useful • ease of use… • reusability…(object oriented) • portable… • TRILINOS http://software.sandia.gov/trilinos • software multi-package…developed at SANDIA (M. Heroux) • modular…no need to install everything in order to work! • Capabilities of LAPACK, AZTEC, Chaco, SuperLU, etc…combined • very active user community…ever evolving! • ease of use…without sacrificing performance Large Scale Computational Challenges in Materials Science

  44. C. Bekas: CS Colloquium Conclusions • Large Scale Challenges in Computational Materials Science • In DFT eigenvalue calculations dominate… • …many O(1000) eigenvalues/vectors required… • …easily reaching and exceeding the limits of state-of-the-art traditional solvers • AMLS appears as an extremely atractive alternative…however accuracy requirements and efficient parallel implementation is still under development Alternative approach: Avoid eigenvalue calculations altogether! • Based on Tchebychev expansions of the Fermi/Dirac distribution… • we design Monte-Carlo simulations for the diagonal of the density matrix… • Required: only Matrix-Vector products with the Hamiltonian… • highly parallelizable! Many open problems in ab initio calculations…one of the most active fields of research today! Large Scale Computational Challenges in Materials Science

More Related