Mercurial > repos > public > sbplib
view +scheme/Euler1d.m @ 1037:2d7ba44340d0 feature/burgers1d
Pass scheme specific parameters as cell array. This will enabale constructDiffOps to be more general. In addition, allow for schemes returning function handles as diffOps, which is currently how non-linear schemes such as Burgers1d are implemented.
author | Vidar Stiernström <vidar.stiernstrom@it.uu.se> |
---|---|
date | Fri, 18 Jan 2019 09:02:02 +0100 |
parents | a35ed1d124d3 |
children | 2b1b944deae1 |
line wrap: on
line source
classdef Euler1d < scheme.Scheme properties m % Number of points in each direction, possibly a vector N % Number of points total h % Grid spacing u % Grid values x % Values of x and y for each order % Order accuracy for the approximation D % non-stabalized scheme operator M % Derivative norm gamma H % Discrete norm Hi e_l, e_r, e_L, e_R; end properties (Constant) SUBSONIC_INFLOW = 1; SUBSONIC_OUTFLOW = -1; NO_FLOW = 0; SUPERSONIC_INFLOW = 2; SUPERSONIC_OUTFLOW = -2; end methods function obj = Euler1d(m,xlim,order,gama,opsGen,do_upwind) default_arg('opsGen',@sbp.D2Standard); default_arg('gama', 1.4); default_arg('do_upwind', false); gamma = gama; [x, h] = util.get_grid(xlim{:},m); if do_upwind ops = sbp.D1Upwind(m,xlim,order); Dp = ops.Dp; Dm = ops.Dm; D1 = (Dp + Dm)/2; Ddisp = (Dp - Dm)/2; else ops = opsGen(m,xlim,order); printExpr('issparse(ops.D1)'); D1 = ops.D1; end H = sparse(ops.H); Hi = sparse(ops.HI); e_l = sparse(ops.e_l); e_r = sparse(ops.e_r); I_x = speye(m); I_3 = speye(3); D1 = kr(D1, I_3); if do_upwind Ddisp = kr(Ddisp,I_3); end % Norms obj.H = kr(H,I_3); obj.Hi = kr(Hi,I_3); % Boundary operators obj.e_l = e_l; obj.e_r = e_r; obj.e_L = kr(e_l,I_3); obj.e_R = kr(e_r,I_3); obj.m = m; obj.h = h; obj.order = order; % Man har Q_t+F_x=0 i 1D Euler, där % q=[rho, rho*u, e]^T % F=[rho*u, rho*u^2+p, (e+p)*u] ^T % p=(gamma-1)*(e-rho*u^2/2); %Solving on form q_t + F_x = 0 function o = Fx(q) Q = reshape(q,3,m); o = reshape(obj.F(Q),3*m,1); o = D1*o; end function o = Fx_disp(q) Q = reshape(q,3,m); f = reshape(obj.F(Q),3*m,1); c = obj.c(Q); lambda_max = c+abs(Q(2,:)./Q(1,:)); % lambda_max = max(lambda_max); lamb_Q(1,:) = lambda_max.*Q(1,:); lamb_Q(2,:) = lambda_max.*Q(2,:); lamb_Q(3,:) = lambda_max.*Q(3,:); lamb_q = reshape(lamb_Q,3*m, 1); o = -D1*f + Ddisp*lamb_q; end if do_upwind obj.D = @Fx_disp; else obj.D = @(q)-Fx(q); end obj.u = x; obj.x = kr(x,ones(3,1)); obj.gamma = gamma; end % Flux function function o = F(obj, Q) % Flux: f = [q2; q2.^2/q1 + p(q); (q3+p(q))*q2/q1]; p = obj.p(Q); o = [Q(2,:); Q(2,:).^2./Q(1,:) + p; (Q(3,:)+p).*Q(2,:)./Q(1,:)]; end % Equation of state function o = p(obj, Q) % Pressure p = (gamma-1)*(q3-q2.^2/q1/2) o = (obj.gamma-1)*( Q(3,:)-1/2*Q(2,:).^2./Q(1,:) ); end % Speed of sound function o = c(obj, Q) % Speed of light c = sqrt(obj.gamma*p/rho); o = sqrt(obj.gamma*obj.p(Q)./Q(1,:)); end % Eigen value matrix function o = Lambda(obj, q) u = q(2)/q(1); c = obj.c(q); L = [u, u+c, u-c]; o = diag(L); end % Diagonalization transformation function o = T(obj, q) % T is the transformation such that A = T*Lambda*inv(T) % where Lambda=diag(u, u+c, u-c) rho = q(1); u = q(2)/q(1); e = q(3); gamma = obj.gamma; c = sqrt(gamma*obj.p(q)/rho); sqrt2gamm = sqrt(2*(gamma-1)); o = [ sqrt2gamm*rho , rho , rho ; sqrt2gamm*rho*u , rho*(u+c) , rho*(u-c) ; sqrt2gamm*rho*u^2/2, e+(gamma-1)*(e-rho*u^2/2)+rho*u*c , e+(gamma-1)*(e-rho*u^2/2)-rho*u*c ; ]; % Devide columns by stuff to get rid of extra comp? end function fs = flowStateL(obj, q) q_l = obj.e_L'*q; c = obj.c(q_l); v = q_l(2,:)/q_l(1,:); if v > c fs = scheme.Euler1d.SUPERSONIC_INFLOW; elseif v > 0 fs = scheme.Euler1d.SUBSONIC_INFLOW; elseif v > -c fs = scheme.Euler1d.SUBSONIC_OUTFLOW; else fs = scheme.Euler1d.SUPERSONIC_OUTFLOW; end end % returns positiv values for inlfow, negative for outflow. % +-1 for subsonic function fs = flowStateR(obj, q) q_r = obj.e_R'*q; c = obj.c(q_r); v = q_r(2,:)/q_r(1,:); if v < -c fs = scheme.Euler1d.SUPERSONIC_INFLOW; elseif v < 0 fs = scheme.Euler1d.SUBSONIC_INFLOW; elseif v < c fs = scheme.Euler1d.SUBSONIC_OUTFLOW; else fs = scheme.Euler1d.SUPERSONIC_OUTFLOW; end end % Enforces the boundary conditions % w+ = R*w- + g(t) function closure = boundary_condition(obj,boundary, type, varargin) [e_s,e_S,s] = obj.get_boundary_ops(boundary); % Boundary condition on form % w_in = R*w_out + g, where g is data switch type case 'wall' closure = obj.boundary_condition_wall(boundary); case 'inflow' closure = obj.boundary_condition_inflow(boundary,varargin{:}); case 'outflow' closure = obj.boundary_condition_outflow(boundary,varargin{:}); case 'inflow_rho' closure = obj.boundary_condition_inflow_rho(boundary,varargin{:}); case 'outflow_rho' closure = obj.boundary_condition_outflow_rho(boundary,varargin{:}); case 'char' closure = obj.boundary_condition_char(boundary,varargin{:}); otherwise error('Unsupported bc type: %s', type); end end % Sets the boundary condition Lq = g, where % L = L(rho, u, e) % p_in are the indecies of the ingoing characteristics. % % Returns closure(q,g) function closure = boundary_condition_L(obj, boundary, L_fun, p_in) [e_s,e_S,s] = obj.get_boundary_ops(boundary); p_ot = 1:3; p_ot(p_in) = []; p = [p_in, p_ot]; % Permutation to sort pt(p) = 1:length(p); % Inverse permutation function o = closure_fun(q,g) % Extract solution at the boundary q_s = e_S'*q; rho = q_s(1); u = q_s(2)/rho; e = q_s(3); c = obj.c(q_s); % Calculate transformation matrix T = obj.T(q_s); Tin = T(:,p_in); Tot = T(:,p_ot); % Calculate eigen value matrix Lambda = obj.Lambda(q_s); % Setup the penalty parameter tau1 = -2*abs(Lambda(p_in,p_in)); tau2 = zeros(length(p_ot),length(p_in)); % Penalty only on ingoing char. tauHat = [tau1; tau2]; tau = e_S*sparse(T*tauHat(pt,:)); L = L_fun(rho,u,e); o = 1/2*obj.Hi * tau * inv(L*Tin)*(L*q_s - g); end closure = @closure_fun; end % Return closure(q,g) function closure = boundary_condition_char(obj,boundary) [e_s,e_S,s] = obj.get_boundary_ops(boundary); function o = closure_fun(q, w_data) q_s = e_S'*q; rho = q_s(1); u = q_s(2)/rho; e = q_s(3); c = obj.c(q_s); Lambda = [u, u+c, u-c]; p_in = find(s*Lambda < 0); p_ot = 1:3; p_ot(p_in) = []; p = [p_in p_ot]; pt(p) = 1:length(p); T = obj.T(q_s); tau1 = -2*diag(abs(Lambda(p_in))); tau2 = zeros(length(p_ot),length(p_in)); % Penalty only on ingoing char. tauHat = [tau1; tau2]; tau = -s*e_S*sparse(T*tauHat(pt,:)); w_s = inv(T)*q_s; w_in = w_s(p_in); w_in_data = w_data(p_in); o = 1/2*obj.Hi * tau * (w_in - w_in_data); end closure = @closure_fun; end % Return closure(q,[v; p]) function closure = boundary_condition_inflow(obj, boundary) [~,~,s] = obj.get_boundary_ops(boundary); switch s case -1 p_in = [1 2]; case 1 p_in = [1 3]; end a = obj.gamma - 1; L = @(rho,u,~)[ 0 1/rho 0; %v 0 -1/2*u*a a; %p ]; closure_raw = boundary_condition_L(obj, boundary, L, g, p_in); closure = @(q,p,v) closure_raw(q,[v; p]); end % Return closure(q, p) function closure = boundary_condition_outflow(obj, boundary) [~,~,s] = obj.get_boundary_ops(boundary); switch s case -1 p_in = 2; case 1 p_in = 3; end a = obj.gamma -1; L = @(~,u,~)a*[0 -1/2*u 1]; closure = boundary_condition_L(obj, boundary, L, p_in); end % Return closure(q,[v; rho]) function closure = boundary_condition_inflow_rho(obj, boundary) [~,~,s] = obj.get_boundary_ops(boundary); switch s case -1 p_in = [1 2]; case 1 p_in = [1 3]; end a = obj.gamma - 1; L = @(rho,~,~)[ 0 1/rho 0; 1 0 0; ]; closure = boundary_condition_L(obj, boundary, L, p_in); end % Return closure(q,rho) function closure = boundary_condition_outflow_rho(obj, boundary) [~,~,s] = obj.get_boundary_ops(boundary); switch s case -1 p_in = 2; case 1 p_in = 3; end L = @(~,~,~)[1 0 0]; closure = boundary_condition_L(obj, boundary, L, p_in); end % Set wall boundary condition v = 0. function closure = boundary_condition_wall(obj,boundary) [e_s,e_S,s] = obj.get_boundary_ops(boundary); % Vill vi sätta penalty på karateristikan som är nära noll också eller vill % vi låta den vara fri? switch s case -1 p_in = 2; p_zero = 1; p_ot = 3; case 1 p_in = 3; p_zero = 1; p_ot = 2; otherwise error(); end p = [p_in, p_zero, p_ot]; % Permutation to sort pt(p) = 1:length(p); % Inverse permutation function o = closure_fun(q) q_s = e_S'*q; rho = q_s(1); u = q_s(2)/rho; c = obj.c(q_s); T = obj.T(q_s); R = -(u-c)/(u+c); % l = [u, u+c, u-c]; % p_in = find(s*l <= 0); % p_ot = find(s*l > 0); tau1 = -2*c; tau2 = [0; 0]; % Penalty only on ingoing char. % Lambda_in = diag(abs(l(p_in))); % Lambda_ot = diag(abs(l(p_ot))); tauHat = [tau1; tau2]; tau = -s*e_S*sparse(T*tauHat(pt)); w_s = inv(T)*q_s; % w_s = T\q_s; % w_s = Tinv * q_s; % Med analytisk matris w_in = w_s(p_in); w_ot = w_s(p_ot); o = 1/2*obj.Hi * tau * (w_in - R*w_ot); end closure = @closure_fun; end function [closure, penalty] = interface(obj, boundary, neighbour_scheme, neighbour_boundary, type) error('NOT DONE') % u denotes the solution in the own domain % v denotes the solution in the neighbour domain [e_u,d1_u,d2_u,d3_u,s_u,gamm_u,delt_u, halfnorm_inv] = obj.get_boundary_ops(boundary); [e_v,d1_v,d2_v,d3_v,s_v,gamm_v,delt_v] = neighbour_scheme.get_boundary_ops(neighbour_boundary); tuning = 2; alpha_u = obj.alpha; alpha_v = neighbour_scheme.alpha; tau1 = ((alpha_u/2)/delt_u + (alpha_v/2)/delt_v)/2*tuning; % tau1 = (alpha_u/2 + alpha_v/2)/(2*delt_u)*tuning; tau4 = s_u*alpha_u/2; sig2 = ((alpha_u/2)/gamm_u + (alpha_v/2)/gamm_v)/2*tuning; sig3 = -s_u*alpha_u/2; phi2 = s_u*1/2; psi1 = -s_u*1/2; tau = tau1*e_u + tau4*d3_u; sig = sig2*d1_u + sig3*d2_u ; phi = phi2*d1_u ; psi = psi1*e_u ; closure = halfnorm_inv*(tau*e_u' + sig*d1_u' + phi*alpha_u*d2_u' + psi*alpha_u*d3_u'); penalty = -halfnorm_inv*(tau*e_v' + sig*d1_v' + phi*alpha_v*d2_v' + psi*alpha_v*d3_v'); end % Ruturns the boundary ops and sign for the boundary specified by the string boundary. % The right boundary is considered the positive boundary function [e,E,s] = get_boundary_ops(obj,boundary) switch boundary case 'l' e = obj.e_l; E = obj.e_L; s = -1; case 'r' e = obj.e_r; E = obj.e_R; s = 1; otherwise error('No such boundary: boundary = %s',boundary); end end function N = size(obj) N = prod(obj.m); end end end