' & $ % Stochastic Methods in Electrostatics: Applications To

' & $ % Stochastic Methods in Electrostatics: Applications To

' $ Stochastic Methods in Electrostatics: Applications to Biological and Physical Science Michael Mascagni a Department of Computer Science and School of Computational Science Florida State University, Tallahassee, FL 32306 USA E-mail: [email protected] URL: http://www.cs.fsu.edu/»mascagni Research supported by ARO, DoD, DOE, NATO, and NSF aWith help from Drs. James Given, Chi-Ok Hwang, and Nikolai Simonov & % ' $ Outline of the Talk First-Passage Algorithms ² Walk on Spheres (WOS) ² Greens Function First Passage (GFFP) ² Simulation-Tabulation (S-T) ² Walk on Subdomains (biochemistry) ² Walk on the Boundary Applications ² Materials Science ² Biochemistry Last-Passage Algorithms Conclusions and Future Work & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 1 of 56 ' $ Stochastic Methods for Partial Di®erential Equations (PDEs) Examples for Solving Elliptic PDEs (Path Integrals) ² Exterior Laplace problems and electrostatics ² Electrical capacitance ² Charge density Advantages of Stochastic Algorithms (Curse of Dimensionality) ² Can avoid complex discrete objects ² Can deal with complicated geometries/interfaces ² Can often cope with singular solutions & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 2 of 56 ' $ Brownian Motion and the Di®usion/Laplace Equations Cauchy problem for the di®usion equation: 1 u = ¢u (1) t 2 u(x; 0) = f(x) (2) in 1-D: Z 1 u(x; t) = !(x ¡ y; t)f(y)dy (3) ¡1 where 2 1 ¡ (x¡y) !(x ¡ y; t) = p e 2t (4) 2¼t & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 3 of 56 ' $ Brownian Motion and the Di®usion/Laplace Equations x u(x; t) = Ex[f(X (t))] (5) ² Xx(t): a Brownian motion which has !(x ¡ y; t) as the transition probability of going from x to y in time t ² Ex[:]: an expectation w.r.t. Brownian motion Z 1 x Ex[f(X (t))] = !(x ¡ y; t)f(y)dy (6) ¡1 & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 4 of 56 ' $ The First Passage (FP) Probability is the Green's Function A related elliptic boundary value problem (Dirichlet problem): ¢u(x) = 0; x 2 ­ u(x) = f(x); x 2 @­ (7) ² Distribution of z is uniform on the sphere ² Mean of the values of u(z) over the sphere is u(x) ² u(x) has mean-value property and harmonic ² Also, u(x) satis¯es the boundary condition x u(x) = Ex[f(X (t@­))] (8) & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 5 of 56 ' $ The First Passage (FP) Probability is the Green's Function Ω @ ª Ü Ü X ´Øµ X ´Ø µ ª z @ first−passage location x; starting point & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 6 of 56 ' $ The First Passage (FP) Probability is the Green's Function (Cont.) Reinterpreting as an average of the boundary values Z u(x) = p(x; y)f(y)dy (9) @­ Another representation in terms of an integral over the boundary Z @g(x; y) u(x) = f(y)dy (10) @­ @n g(x; y) { Green's function of the Dirichlet problem in ­ @g(x; y) =) p(x; y) = (11) @n & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 7 of 56 ' $ `Walk on Spheres' (WOS) and Green's Function First Passage (GFFP) Algorithms ² Green's function is known =) direct simulation of exit points and computation of the solution through averaging boundary values ² Green's function is unknown =) simulation of exit points from standard subdomains of ­, e.g. spheres =) Markov chain of `Walk on Spheres' (or GFFP algorithm) fx0 = x; x1;:::g xi ! @­ and hits "-shell is N = O(ln(")) steps xN simulates exit point from ­ with O(") accuracy & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 8 of 56 ' $ `Walk on Spheres' (WOS) and Green's Function First Passage (GFFP) Algorithms WOS: O Shell thickness ε & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 9 of 56 ' $ Timing of the 'Walk on Spheres' Algorithm 3500 3000 2500 2000 running time (secs) 1500 1000 logarithmic regression 500 10−8 10−7 10−6 10−5 10−4 10−3 10−2 10−1 100 ε & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 10 of 56 ' $ Various Laplacian Green's Functions: GFFP O O O (a) Putting back (b) Void space (c)Intersecting & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 11 of 56 ' $ Geometry for Permeability Computations & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 12 of 56 ' $ The Simulation-Tabulation (S-T) Method for Generalization ² Green's function for the non-intersected surface of a sphere located on the surface of a reflecting sphere Absorbing Sphere Ω 2 r 2 Ω 1 O2 r 1 Reflecting Sphere O1 & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 13 of 56 ' $ Example: Solc-Stockmayer Model without Potential Reactive patch θ R R0 & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 14 of 56 ' $ Another S-T Application: Mean Trapping Rate In a domain of nonoverlapping spherical traps : δ O & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 15 of 56 ' $ Biological Electrostatics: Motivation Electrostatics are Extremely Important in Quantitative Biochemistry: ² Ligand binding ² Protein-protein interactions ² Protein-nucleic acid interactions ² Prediction from primary structure information In Vivo Electrostatics Must Include the Solvent ² Explicit solvent model: individual water/ions computed, often with Molecular Dynamics ² Implicit solvent models: continuum model of water and dissolved ions used, Poisson-Boltzmann equation & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 16 of 56 ' $ Molecular Electrostatics Problem Implicit solvent model ² Poisson equation for the electrostatic potential, Á: ¡r²(x)rÁ(x) = 4¼½(x) ; x 2 R3 dielectric permittivity, ², and charge density, ½, are position-dependent 3 ² molecule ­ { a compact cavity in R with low ² = ²i ² surrounded by solvent with larger ² = ²e ² point charges, qm, at xm inside molecule ² Boltzmann distribution of mobile ions in solvent & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 17 of 56 ' $ Molecular Electrostatics Problem (Cont.) Explicit geometric models for solute molecule ² van der Waals surface: SM (m) (m) union of intersecting spheres (atoms): ­ = m=1 B(x ; r ) (m) point charges { at their centers, xm = x ² contact and reentrant surface, ¡: @­ smoothed by the probe molecule of the solute rolling on it ² ion-accessible surface @­0: 0 SM (m) (m) ­ = m=1 B(x ; r + rion); ion-exclusion layer between @­0 and ¡ & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 18 of 56 ' $ Molecular Electrostatics Problem (Cont.) Mathematical model (one-surface geometry) ² Poisson equation for the electrostatic potential, Á, inside a molecule XM ¡²i¢Ái(x) = 4¼qm±(x ¡ xm) ; x 2 ­ m=1 ² linearized Poisson-Boltzmann equation outside, x 2 R3 n ­: 2 ¢Áe(x) ¡ · Áe(x) = 0 ; ² Continuity condition on the boundary @Á @Á Á = Á ; ² i = ² e ; y 2 ¡ ´ @­ i e i @n(y) e @n(y) & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 19 of 56 ' $ Electrostatic Potential, Field and Energy (Linear Problem) ² Point values of the potential: Á(x) = Á(0)(x) + g(x) Here, singular part of Á: XM q 1 g(x) = m ² jx ¡ x j m=1 i m ² Free electrostatic energy of a molecule = linear combination of point values of the regular part of the electrostatic potential Á(0): 1 XM E = Á(0)(x )q ; 2 m m m=1 ² Point values of the electrostatic ¯eld: rÁ(x) & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 20 of 56 ' $ Monte Carlo Estimates for Point Potential Values Two di®erent approaches to constructing Monte Carlo algorithms 1. Probabilistic representation for the solution 2. Classical potential theory First approach Laplace equation for the regular part of the potential inside ­ ¢Á(0) = 0 Probabilistic representation (0) (0) ¤ Á (x) = Ex[Á (x )] x¤ { exit point from ­ of Brownian motion starting at x & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 21 of 56 ' $ `Walk on Spheres' Algorithm x { center of a sphere ) exit points are distributed isotropically. Ball B(x; R) lies entirely in ­. Strong Markov property of Brownian motion ) probabilistic representation holds valid for exit points. Hence follows `random walk on spheres' algorithm for general domains with regular boundary: xk = xk¡1 + d(xk¡1) £ !k ; k = 1; 2;:::: Here d(xk¡1) { distance from xk¡1 to the boundary f!kg { sequence of independent unit isotropic vectors xk is exit point from the ball, B(xk¡1; d(xk¡1)), for Brownian motion starting at xk¡1 & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 22 of 56 ' $ Green's Function First Passage Simulation Other domains with known Green's function (G) () one-step simulation of exit points distributed on the boundary in accordance with @G=@n For general domains: E±cient way to simulate x¤ { combination of `walk in subdomains' approach and `walk on spheres' algorithm The whole domain, ­, is represented as a union of intersecting subdomains: [M ­ = ­m m=1 Simulate exit point separately in every ­m & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 23 of 56 ' $ Green's Function First Passage Simulation (cont.) x0 = x; x1; : : : ; xN { Markov chain, every xi+1 is exit point from the corresponding subdomain for Brownian motion starting at xi i i For spherical subdomains, B(xm;Rm), exit points are distributed in accordance with the Poisson's kernel i i i i 2 i jx ¡ xmj jx ¡ xmj ¡ Rm i i 3 4¼Rm jx ¡ yj x¤ = xN is exit point of Brownian motion from ­ Schwartz lemma ) Markov chain fxig converges to x¤ geometrically & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 24 of 56 ' $ `Walk on Spheres' and `Walk in Subdomains' Algorithms Figure 1: Walk in subdomains example. & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 25 of 56 ' $ Monte Carlo Estimate for Point Potential Value On every step Á(0)(xi) = E[Á(0)(xi+1)jxi] Hence Á(0)(x) = EÁ(0)(x¤) ´ E[Á(x¤) ¡ g(x¤)] Values of the electrostatic potential on the boundary, Á(x¤), are not known. We can use their Monte Carlo estimates instead & % Prof. Michael Mascagni: Stochastic Electrostatics Slide 26 of 56 ' $ Monte Carlo Estimate for Boundary Potential Values ² First approach Discretization and randomization of the boundary condition (y 2 ¡, n = n(y) { normal vector); 2 Á(y) = piÁ(y ¡ hn) + peÁ(y + hn) + O(h ) =) ¤ 0 ¤ 2 Á(x1) = E(Á(x2jx1) + O(h ) 0 ¤ x2 = x1 ¡ hn with probability pi (reenter molecule) 0 ¤ x2 = x1 + hn with probability pe = 1 ¡ pi (exit to solvent) ²i pi = ²i + ²e & % Prof.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    57 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us