keyword
stringclasses
7 values
repo_name
stringlengths
8
98
file_path
stringlengths
4
244
file_extension
stringclasses
29 values
file_size
int64
0
84.1M
line_count
int64
0
1.6M
content
stringlengths
1
84.1M
language
stringclasses
14 values
2D
pillowlab/GLMspiketools
glmtools_spline/neglogli_tspline_poiss.m
.m
1,247
49
function [L,dL,ddL] = neglogli_tspline_poiss(prs,X,Y,g,dtbin) % [L,dL,ddL] = neglogli_tspline_poiss(prs,X,Y,g,dtbin) % % Compute negative log-likelihood of data Y given rate g(X*prs). % % INPUT: % prs [M x 1] - parameters % X [N x M] - design matrix % Y [N x 1] - observed Poisson random variables % g - handle for transfer function % dtbin [1 x 1] - size of time bins of Y % % OUTPUT: % L [1 x 1] - negative log-likelihood % dL [M x 1] - gradient % ddL [M x M] - Hessian (2nd deriv matrix) % % Last modified: 25 May 2018 (JW Pillow) % Project parameters onto design matrix z = X*prs; etol = 1e-100; if nargout==1 % Compute neglogli f = g(z); f(f<etol)=etol; L = -Y'*log(f) + sum(f)*dtbin; elseif nargout == 2 % Compute neglogli & Gradient [f,df] = g(z); f(f<etol)=etol; L = -Y'*log(f) + sum(f)*dtbin; % grad wts = (dtbin*df-(Y.*df./f)); dL = X'*wts; elseif nargout == 3 % Compute neglogli, Gradient & Hessian [f,df,ddf] = g(z); f(f<etol)=etol; L = -Y'*log(f) + sum(f)*dtbin; % grad wts = (dtbin*df-(Y.*df./f)); dL = X'*wts; ww = dtbin*ddf-Y.*(ddf./f-(df./f).^2); ddL = X'*bsxfun(@times,X,ww); end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_spline/MLfit_splineNlin.m
.m
985
30
function [gg,negloglival,pp] = MLfit_splineNlin(gg,Stim) % [gg,neglogli,pp] = MLfit_splineNlin(gg,Stim,optimargs) % % Fit a nonlinear function in an GLM neuron with a (positive) spline defined using a set of knots % (discontinuities of a piecewise-polynomial function), using maximum likelihood % loss % % Inputs: % gg = glm param structure % Stim = stimulus % % Outputs: % fun - function pointer to nonlinearity (uses 'ppval') % pp - piecewise polynomial structure % Mspline - matrix for converting function params to spline coeffs % splinePrs - coefs=Mspline*splinePrse % Compute net GLM filter output [~,~,~,Itot] = neglogli_GLM(gg,Stim); Itot = Itot-gg.dc; % remove effect of dc gg.dc = 0; % set dc to zero % Do fitting [fspline,pp,negloglival] = fit_tspline_poisson(Itot,gg.sps,gg.splineprs,gg.dtSp); % Insert params into glm struct gg.ppstruct = pp; % piecewise polynomial parameters gg.nlfun = fspline; % insert nonlinearity
MATLAB
2D
pillowlab/GLMspiketools
glmtools_spline/fminNewton.m
.m
5,145
181
function [prs,fval,H] = fminNewton(fptr,prs,opts) % [prs,fval,H] = fminNewton(fptr,prs,opts) % % Simple mplementation of Newton's method for minimizing a function % % Inputs: % fptr - function handle for loss function % prs - initial value of parameters % opts - struct with fields: 'tolX', 'tolFun', 'maxIter', 'verbose'. % (verbose takes on values 0, 1 or 2) % % Outputs: % prs - minimizer of function (or best guess) % fval - value of function at minimizer % H - Hessian at minimizer % % Derivation of Newton's method (refresher): % ------------------------------ % f = f0 + f1(x-x0) + 1/2 (x-x0)^T H (x-x0) %% 2nd order Taylor expansion % df = f1 + H(x-x0) %% derivative % x = x0 - H^{-1} f1 %% solve for x (new point) % ------------------------------ % % $Id$ if nargin < 3 opts = []; end % set options to defaults, if necessary if isfield(opts,'tolX'),tolX= opts.tolX; else tolX = 1e-8; end if isfield(opts,'tolFun'),tolFun= opts.tolFun; else tolFun = 1e-8; end if isfield(opts,'maxIter'),maxIter= opts.maxIter; else maxIter = 50; end if isfield(opts,'verbose'),verbose= opts.verbose; else verbose = 0; end % Initialize dX = inf; dF = inf; iter = 0; canDescend = true; nprs = length(prs); % % Change warning states so ill-conditioned matrix inverses aren't reported % msgid1 = 'MATLAB:illConditionedMatrix'; % msgid2 = 'MATLAB:singularMatrix'; % msgid3 = 'MATLAB:nearlySingularMatrix'; % wstate = warning('off',msgid1); % warning('off',msgid2); % warning('off',msgid3); % lastwarn(''); while (dX>tolX) && (dF>tolFun) && (iter < maxIter) && canDescend % evaluate function, gradient & Hessian [f,df,H] = fptr(prs); if any(isinf(H)) | any(isnan(H)) error('fminNewton2: Infinity or Nan in user-supplied Hessian'); end % Test for positive definite-ness try chol(H); REG=0; catch fprintf('[[not pos-def]]');REG=1; end % Check to see if H is ill-conditioned if ~REG && (cond(H)<1e6) % ---- Standard Newton's Method if well-conditioned ---- dprs = -H\df; % % Check if it's a descent direction % if dprs'*df > 0 % dprs = -dprs; % flip sign otherwise % fprintf(1, 'FLIPPED\n'); % end else % ---- Regularize if ill-conditioned ---- if verbose>=2 fprintf('(regularizing H)'); end % Compute eigenvalues of H eigvals = eig(H); eigmin = min(eigvals);eigmax = max(eigvals); if eigmin > 0 lam = eigmax/1e3; else %lam = max(eigmax/1e6,-2*eigmin); lam = eigmax/1e3-eigmin; end dprs = -(H+lam*speye(nprs))\(df); % next attempt % Make sure *this* is a descent direction (if not, something is wrong) if dprs'*df > 0 error('Somehow we didn''t end up with a descent direction'); end end % Evaluate function at new value fnew = fptr(prs+dprs); if fnew < f % Accept the position: update bookkeeping params dX = norm(dprs); dF = -(fnew-f); % Update params prs = prs+dprs; else % Line search in this search direction % (Note: will give errs if loss func doesn't accept matrix inputs) if verbose >= 2 fprintf('(line search 1)'); end fracs = 3.^(-(1:5)); % fractions of gradient to move fnewvec = fptr(bsxfun(@plus,prs,+dprs*fracs)); [fnew,imin] = nanmin(fnewvec); if ~(fnew<f) % try again with even smaller steps if verbose >= 2 fprintf('(line search 2)'); end fracs = 3.^(-(6:15)); % fractions of gradient to move fnewvec = fptr(bsxfun(@plus,prs,+dprs*fracs)); [fnew,imin] = nanmin(fnewvec); end if isinf(fnew) || isnan(fnew) error('fminNewton2: unavoidable infinity or Nan function value: try improving line search'); end if ~(fnew<f) canDescend = 0; if verbose>=1 fprintf('fminNewton: can''t descend anymore; minimum possible\n'); end else % Accept the position: update bookkeeping params dprs = dprs*fracs(imin); dX = norm(dprs); dF = -(fnew-f); % Update params prs = prs+dprs; end end % Print out information if requested if verbose>=2 fprintf('iter %d: fun=%.4f\n', iter,fnew); end iter = iter+1; % % Check to make sure we're only decreasing function % if canDescend && (dF<0) % warning('error in fminNewton: function increased somehow!'); % for debugging % keyboard; % end end % If desired, Compute value and Hessian at final value if nargout > 1 [fval,~,H] = fptr(prs); end % Report stopping condition, if verbose >= 1 if verbose >=1 if ~canDescend && iter>3 fprintf('fminNewton could not descend further (iter=%d): flat local minimum likely\n',iter); elseif ~canDescend fprintf('fminNewton could not descend (iter=%d): quadratic approx may be bad\n',iter); fprintf('Try implementing a linesearch method or using better initialization\n'); elseif dX<tolX fprintf('fminNewton finished (iter=%d): change in X less than tolX\n',iter); elseif dF<tolFun fprintf('fminNewton finished (iter=%d): change in fun less than tolFun\n',iter); else fprintf('fminNewton stopped: iter=%d exceeded maxIter',iter); end end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_spline/tsplinefun.m
.m
483
24
function [f,df,ddf] = tsplinefun(x,splfun,tfun) % [f,df,ddf] = tsplinefun(x,splfun,tfun) % % Computes the function: % f(x) = tfun(splfun(x)) and its first and second derivatives % where tfun is a function handle switch nargout case 0, f = tfun(splfun(x)); case 1, f = tfun(splfun(x)); case 2, [g,dg] = splfun(x); [f,df] = tfun(g); df = df.*dg; case 3, [g,dg,ddg] = splfun(x); [f,dfg,ddfg] = tfun(g); df = dfg.*dg; ddf = ddfg.*dg.^2 + dfg.*ddg; end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_spline/initializeGLMsplineNlin.m
.m
1,847
45
function [ggnew,negloglival] = initializeGLMsplineNlin(gg,Stim,prs) % [ggnew,negloglival] = initializeGLMsplineNlin(gg,Stim,prs) % % Initializes parameters for a spline nonlinearity using GLM log-likelihood % % Inputs: % gg = param struct % Stim = stimulus % spline_prs = structure parameters for spline nonlinearity % ---- Fields: ---- % nknots = # of knots, % epprob = [low,high] quantiles for first and last knots % smoothness = 1+degree of smoothness (3 -> 2nd order continuous) % extrapDeg = [lo,hi] degree of polynomials for extrapolation outside breaks % % Outputs: % ggnew = new param struct (with ML params); % negloglival = negative log-likelihood at ML estimate % Compute net current output from GLM [~,~,~,Itot] = neglogli_GLM(gg,Stim); Itot = Itot-gg.dc; % remove DC from injected current % Set knots and insert piece-wise linear nonlinearity. knotrange = quantile(Itot,prs.epprob); knots = linspace(knotrange(1),knotrange(2), prs.nknots); prebreaks = ((prs.extrapDeg(1):1)-2)/10; % extra left breaks (spaced 0.1 apart) postbreaks = (1:(2-prs.extrapDeg(2)))/10; % extra right breaks (spaced 0.1 apart) breaks = [prebreaks+knots(1), knots, knots(end)+postbreaks]; % make spline structure ss = struct('breaks',breaks, 'smoothness',prs.smoothness, 'extrapDeg',prs.extrapDeg,'tfun',@logexp1); [fspline,pp,negloglival] = fit_tspline_poisson(Itot,gg.sps,ss,gg.dtSp); % Insert params into glm struct ggnew = gg; % initialize ggnew.dc = 0; % set DC term to zero ggnew.ppstruct = pp; % piecewise polynomial parameters ggnew.splineprs = ss; % spline hyperparameters ggnew.nlfun = fspline; % insert nonlinearity % % Optional: make plot comparing old and new nonlinearity % xx = linspace(min(II),max(II),100); % plot(xx,fspline(xx),breaks,fspline(breaks),'o',xx,gg.nlfun(xx));
MATLAB
2D
pillowlab/GLMspiketools
glmtools_spline/fit_tspline_poisson.m
.m
2,410
66
function [fun,pp,neglogli,splPrs,Mspline] = fit_tspline_poisson(x,y,ss,dtbin,prs0) % [fun,pp,neglogli,splPrs,Mspline] = fit_tspline_poisson(x,y,ss,dtbin,prs0) % % Fit a function y = f(x) with a cubic spline, defined using a set of % breaks, smoothness and extrapolation criteria, by maximizing Poisson % likelihood: y ~ Poiss(g(spline(x))). % % Inputs: % x [Nx1] - input variable % y [Nx1] - output (count) variable % ss [struct] - spline struct with fields: % .breaks - breaks between polynomial pieces % .smoothness - derivs of 1 less are continuous % (e.g., smoothness=3 -> 2nd derivs are continuous) % .extrapDeg - degree polynomial for extrapolation on ends % .tfun - nonlinear transfer function (forcing positive outputs) % dtbin [1x1]- time bin size (OPTIONAL; assumed 1 otherwise) % prs0 [Mx1] - initial guess at spline params (OPTIONAL) % % Outputs: % fun - function handle for nonlinearity (uses 'ppval') % pp - piecewise polynomial structure % neglogli - negative loglikelihood at parameter estimate % splinePrs - vector x such that spline coeffs = Mspline*x % Mspline - matrix for converting params to spline coeffs % % Note: the fitted nonlinearity can be evaluated via: % y = ss.tfun(ppval(pp,xx)); % % last updated: 26/03/2012 (JW Pillow) if nargin<4 dtbin=1; end sortflag=1; % allow sorted design matrix (slightly faster) % Compute design matrix [Xdesign,Ysrt,Mspline] = mksplineDesignMat(x,y,ss,sortflag); nprs = size(Xdesign,2); if nargin<5 prs0 = randn(nprs,1)*.1; end % Set up loss function floss = @(prs)neglogli_tspline_poiss(prs,Xdesign,Ysrt,ss.tfun,dtbin); % HessCheck(floss,prs0); % Check that analytic grad and Hessian are correct % minimize negative log-likelihood using Newton's method [splPrs,neglogli] = fminNewton(floss,prs0); % Create function handle for resulting tspline [~,pp] = mksplinefun(ss.breaks, Mspline*splPrs); splfun = @(x)ppfun(pp,x); tfun = ss.tfun; fun = @(x)tsplinefun(x,splfun,tfun); % % ----------------------------------------- % % If desired, use Matlab's fminunc instead: % opts = optimset('display','off','gradobj','on','Hessian','off','maxiter',5000,'maxfunevals',5000,'largescale','off'); % splinePrs2 = fminunc(floss,prs0,opts); % [floss(splinePrs) floss(splinePrs2)] % Compare results
MATLAB
2D
pillowlab/GLMspiketools
glmtools_spline/mksplinefun.m
.m
544
19
function [fun,pp] = mksplinefun(breaks,x) % [fun,pp] = mksplinefun(breaks,x) % % Make spline function handle out of breaks (knots) and coefficients in vector x. % % Inputs: % breaks - points of discontinuity % x - vector that can be reshaped into N x 4 matrix of spline coefficients % for each segment % % Outputs: % fun - function pointer for spline function % pp - matlab piecewise polynomial structure (evaluate with 'ppval'). coeffs = reshape(x,4,[])'; pp = mkpp(breaks, coeffs); fun = @(x)ppval(pp,x);
MATLAB
2D
pillowlab/GLMspiketools
nlfuns/logexp4.m
.m
559
24
function [f,df,ddf] = logexp4(x); % [f,df,ddf] = logexp4(x); % % Implements the nonlinearity: % f(x) = log(1+exp(x)).^4; % plus first and second derivatives % % General formulas: % f(x) = log(1+exp(x))^k % f'(x) = k log(1+e^x)^(k-1) * e^x/(1+e^x); % f"(x) = k(k-1) log(1+e^x)^(k-2) * (e^x/(1+e^x))^2 % + k log(1+e^x)^(k-1) * e^x/(1+e^x)^2 f0 = log(1+exp(x)); f = f0.^4; if nargout > 1 df = 4*f0.^3.*exp(x)./(1+exp(x)); end if nargout > 2 ddf = 12*f0.^2.*(exp(x)./(1+exp(x))).^2 + ... 4*f0.^3.*exp(x)./(1+exp(x)).^2; end
MATLAB
2D
pillowlab/GLMspiketools
nlfuns/logexp_pow.m
.m
508
23
function [f,df,ddf] = logexp_pow(x,pow); % [f,df,ddf] = logexp_pow(x); % % Implements the nonlinearity: % f(x) = log(1+exp(x)).^pow; % plus first and second derivatives f0 = log(1+exp(x)); f = f0.^pow; if nargout > 1 df = pow*f0.^(pow-1).*exp(x)./(1+exp(x)); end if nargout > 2 if pow == 1 ddf = pow*f0.^(pow-1).*exp(x)./(1+exp(x)).^2; else ddf = pow*f0.^(pow-1).*exp(x)./(1+exp(x)).^2 + ... pow*(pow-1)*f0.^(pow-2).*(exp(x)./(1+exp(x))).^2; end end
MATLAB
2D
pillowlab/GLMspiketools
nlfuns/logexp1.m
.m
368
22
function [f,df,ddf] = logexp1(x); % [f,df,ddf] = logexp1(x); % % Implements the nonlinearity: % f(x) = log(1+exp(x)).^pow; % Where pow = 1; % plus first and second derivatives % pow = 1; f0 = log(1+exp(x)); f = f0.^pow; if nargout > 1 df = pow*f0.^(pow-1).*exp(x)./(1+exp(x)); end if nargout > 2 ddf = pow*f0.^(pow-1).*exp(x)./(1+exp(x)).^2; end
MATLAB
2D
pillowlab/GLMspiketools
nlfuns/logexp2.m
.m
430
20
function [f,df,ddf] = logexp2(x); % [f,df,ddf] = logexp_pow(x); % % Implements the nonlinearity: % f(x) = log(1+exp(x)).^pow; % where pow=2, % plus first and second derivatives pow = 2; f0 = log(1+exp(x)); f = f0.^pow; if nargout > 1 df = pow*f0.^(pow-1).*exp(x)./(1+exp(x)); end if nargout > 2 ddf = pow*f0.^(pow-1).*exp(x)./(1+exp(x)).^2 + ... pow*(pow-1)*f0.^(pow-2).*(exp(x)./(1+exp(x))).^2; end
MATLAB
2D
pillowlab/GLMspiketools
nlfuns/logexp3.m
.m
358
19
function [f,df,ddf] = logexp3(x); % [f,df,ddf] = logexp3(x); % % Implements the nonlinearity: % f(x) = log(1+exp(x)).^3; % plus first and second derivatives f0 = log(1+exp(x)); f = f0.^3; if nargout > 1 df = 3*f0.^2.*exp(x)./(1+exp(x)); end if nargout > 2 ddf = 6*f0.*(exp(x)./(1+exp(x))).^2 + ... 3*f0.^2.*exp(x)./(1+exp(x)).^2; end
MATLAB
2D
pillowlab/GLMspiketools
nlfuns/expfun.m
.m
166
9
function [f,df,ddf] = expfun(x) % [f,df,ddf] = expfun(x) % % replacement for 'exp' that returns 3 arguments (value, 1st & 2nd deriv) f = exp(x); df = f; ddf = df;
MATLAB
2D
pillowlab/GLMspiketools
nlfuns/logexp5.m
.m
559
24
function [f,df,ddf] = logexp5(x); % [f,df,ddf] = logexp4(x); % % Implements the nonlinearity: % f(x) = log(1+exp(x)).^4; % plus first and second derivatives % % General formulas: % f(x) = log(1+exp(x))^k % f'(x) = k log(1+e^x)^(k-1) * e^x/(1+e^x); % f"(x) = k(k-1) log(1+e^x)^(k-2) * (e^x/(1+e^x))^2 % + k log(1+e^x)^(k-1) * e^x/(1+e^x)^2 f0 = log(1+exp(x)); f = f0.^5; if nargout > 1 df = 5*f0.^4.*exp(x)./(1+exp(x)); end if nargout > 2 ddf = 20*f0.^3.*(exp(x)./(1+exp(x))).^2 + ... 5*f0.^4.*exp(x)./(1+exp(x)).^2; end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/normalizecols.m
.m
171
7
function B = normalizecols(A); % % B = normalizecols(A); % % Normalizes the columns of a matrix, so each is a unit vector. B = A./repmat(sqrt(sum(A.^2)), size(A,1), 1);
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/spikefilt.m
.m
697
24
function Y = spikefilt(sps, H) % Y = spikefilt(sps, H, twin) % % Causally filters spike train with spike indices spikeInds with filter matrix H. % % Inputs: % sps [tlen x 1] - spike train vector % H [n x m] - spike-history matrix (each column is a filter) % % Output: % Y [tlen x m] - each column is the spike train filtered with the % corresponding column of H % % Note: there is a faster 'mex' version of this function called spikefilt_mex, % available in version 'v1'. It was removed in Jan 2016 to allow for % mex-free code release. [hlen,hwid] = size(H); % Do convolution and remove extra bins Y = conv2(sps,H,'full'); Y = [zeros(1,hwid); Y(1:end-hlen,:)];
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/makeStimRows.m
.m
2,036
67
function S = makeStimRows(Stim, nkt, flag) % S = makeStimRows(Stim, nkt, flag); % % Converts spatio-temporal stimulus to a design matrix, where each row of % the design matrix contains all space-time stimulus elements within a time % window of "nkt" time bins. % % INPUT: % Stim [N x M] - stimulus, first dimension is time, others are space % nkt [1 x 1] - number of time bins to include in temporal kernel % flag (optional) % 'same' - padded w/ zeros to make S the same size (default) % 'valid' - no padding with zeros: length(S) = length(Stim)-nkt+1. % vector of indices - returns only stimuli corresponding to % these time bins % % OUTPUT: % S [N x M*nkt] or [Nsp x M*nkt ] - design matrix % % $Id$ % parse inputs if nargin < 3 flag = 'same'; end [n1,n2] = size(Stim); if ischar(flag) % --- Form for all stimuli (with zero-padding of first bins) ----- if strcmp(flag,'same') S = zeros(n1, n2*nkt); Stim = [zeros(nkt-1,n2); Stim]; n1 = n1+nkt-1; elseif strcmp(flag,'valid') S = zeros(n1-nkt+1,n2*nkt); else error('Unknown flag'); end % Form design matrix for j=1:n2 % Loop over columns of original stimulus S(:,(nkt*(j-1)+1):(nkt*j)) = ... fliplr(toeplitz(Stim(nkt:n1, j), Stim(nkt:-1:1,j))); end else % ---- Form for spiking stimuli only ---------------------------------------- inds = flag; % indices to keep if (min(inds) < 1) || (max(inds) > n1) error('Spike indices outside allowed range'); end S = zeros(length(inds), n2*nkt); % Do for spinds < nkt nsp1 = length(inds(inds<nkt)); for j = 1:nsp1 S(j,:) = reshape([zeros(nkt-inds(j), n2); Stim(1:inds(j),:)], 1, n2*nkt); end % Do for remaining indices for j = nsp1 +1:length(inds) S(j,:) = reshape(Stim(inds(j)-nkt+1:inds(j),:), 1, n2*nkt); end end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/getFminOptsForVersion.m
.m
736
18
function opts = getFminOptsForVersion(v) % opts = getFminOptsForVersion(v) % % Gets the options for fminunc needed for user's version of MATLAB % % Note of complaint: it's stupid that I had to write this function, but % matlab has incompatible syntax for fminunc options for different versions % (earlier versions of matlab will crash with options required by later % versions). The cutoff point for my two versions was 9.1 vs. 9.3, but if % you're getting warnings or crashes, please report to pillow@princeton.edu % or make an issue report on github if str2num(v(1:3))>9.2 % check if version is before V 9.2 opts = {'algorithm','trust-region','Gradobj','on','Hessian','on'}; else opts = {'Gradobj','on','Hessian','on'}; end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/in.m
.m
461
22
function [ii, nds] = in(v, rnge) % [ii,nds] = in(v, rnge) % % returns indices of vector v within range rnge, inclusive of endpoints if (nargout == 1) if length(rnge) == 1; ii = v(v<=vrnge); else ii = v((v>=rnge(1)) & (v<=rnge(2))); end else if length(rnge) == 1; nds = find(v<=vrnge); ii = v(nds); else nds = find((v>=rnge(1)) & (v<=rnge(2))); ii = v(nds); end end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/simGLM.m
.m
7,517
185
function [tsp,sps,Itot,Istm] = simGLM(glmprs,Stim) % [tsp,sps,Itot,Ispk] = simGLM(glmprs,Stim) % % Compute response of glm to stimulus Stim. % % Uses time rescaling instead of Bernouli approximation to conditionally % Poisson process % % Dynamics: Filters the Stimulus with glmprs.k, passes this through a % nonlinearity to obtain the point-process conditional intensity. Add a % post-spike current to the linear input after every spike. % % Input: % glmprs - struct with GLM params, has fields 'k', 'h','dc' for params % and 'dtStim', 'dtSp' for discrete time bin size for stimulus % and spike train (in s). % Stim - stimulus matrix, with time running vertically and each % column corresponding to a different pixel / regressor. % Output: % tsp - list of spike times (in s) % sps - binary matrix with spike times (at resolution dtSp). % Itot - summed filter outputs % Istm - just the spike-history filter output % Check nonlinearity (default is exponential) if ~isfield(glmprs, 'nlfun'), glmprs.nlfun = @exp; end upSampFactor = glmprs.dtStim/glmprs.dtSp; % number of spike bins per Stim bin assert(mod(upSampFactor,1) == 0, 'dtStim / dtSp must be an integer'); % Determine which version to run if size(glmprs.k,3) > 1 % Run "coupled" GLM model if multiple cells present [tsp,sps,Itot,Istm] = simGLMcpl(glmprs,Stim,upSampFactor); else % Single cell simulation [tsp,sps,Itot,Istm] = simGLMsingle(glmprs,Stim,upSampFactor); end % ====================================================================== function [tsp,sps,Itot,Istm] = simGLMsingle(glmprs,Stim,upSampFactor) % Sub-function within simGLM.m % % Simulates the GLM point process model for a single (uncoupled) neuron % using time-rescaling % --------------- Check Inputs ---------------------------------- nbinsPerEval = 100; % Default number of bins to update for each spike dt = glmprs.dtSp; % bin size for simulation slen = size(Stim,1); % length of stimulus rlen = slen*upSampFactor; % length of binned spike response hlen = size(glmprs.ih,1); % length of post-spike filter % ------------- Compute filtered resp to signal ---------------- Istm = sameconv(Stim,glmprs.k); % filter stimulus with k Istm = kron(Istm,ones(upSampFactor,1)) + glmprs.dc; % upsample stim current Itot = Istm; % total filter output % --------------- Set up simulation dynamics variables --------------- nsp = 0; % number of spikes sps = zeros(rlen,1); % sparse spike time matrix jbin = 1; % current time bin tspnext = exprnd(1); % time of next spike (in rescaled time) rprev = 0; % Integrated rescaled time up to current point % --------------- Run dynamics --------------------------------------- while jbin <= rlen iinxt = jbin:min(jbin+nbinsPerEval-1,rlen); rrnxt = glmprs.nlfun(Itot(iinxt))*dt; % Cond Intensity rrcum = cumsum(rrnxt)+rprev; % integrated cond intensity if (tspnext >= rrcum(end)) % No spike in this window jbin = iinxt(end)+1; rprev = rrcum(end); else % Spike! ispk = iinxt(find(rrcum>=tspnext, 1, 'first')); % time bin where spike occurred nsp = nsp+1; sps(ispk) = 1; % spike time mxi = min(rlen, ispk+hlen); % max time affected by post-spike kernel iiPostSpk = ispk+1:mxi; % time bins affected by post-spike kernel if ~isempty(iiPostSpk) Itot(iiPostSpk) = Itot(iiPostSpk)+glmprs.ih(1:mxi-ispk); end tspnext = exprnd(1); % draw next spike time rprev = 0; % reset integrated intensity jbin = ispk+1; % Move to next bin % -- Update # of samples per iter --- muISI = jbin/nsp; nbinsPerEval = max(20, round(1.5*muISI)); end end tsp = find(sps>0)*dt; % ======================================================================== function [tsp,sps,Itot,Istm] = simGLMcpl(glmprs,Stim,upSampFactor) % [tsp,sps,Itot,Istm] = simGLMcpl(glmprs,Stim,upSampFactor) % % Compute response of (multi-cell) coupled-glm to stimulus Stim. % % Uses time rescaling to sample conditionally Poisson process % % Dynamics: Filters the Stimulus with glmprs.k, passes this through a % nonlinearity to obtain the point-process conditional intensity. Add a % post-spike current to the linear input after every spike. % % Sub-function within simGLM.m % --------------- Check Inputs ---------------------------------- ncells = size(glmprs.k,3); nbinsPerEval = 100; % Default number of bins to update for each spike dt = glmprs.dtSp; % Check that dt evenly divides 1 if mod(1,dt) ~= 0 dt = 1/round(1/dt); fprintf(1, 'glmrunmod: reset dtsim = %.3f\n', dt); end slen = size(Stim,1); % length of stimulus rlen = slen*upSampFactor; % length of binned response ih = permute(glmprs.ih,[1 3 2]); % permute so all outgoing filters in a single plane hlen = size(ih,1); % length of h filter % ------------- Compute filtered resp to signal ---------------- Istm = zeros(slen,ncells); for jj = 1:ncells Istm(:,jj) = sameconv(Stim,glmprs.k(:,:,jj)); % filter stimulus with k end Istm = bsxfun(@plus,Istm,glmprs.dc); % add DC term % Compute finely sampled version Istm = kron(Istm,ones(upSampFactor,1)); % upsample stim current Itot = Istm; % total filter output % --------------- Set up simulation dynamics variables --------------- tsp(1,1:ncells) = {zeros(round(slen/25),1)}; % allocate space for spike times nsp = zeros(1,ncells); jbin = 1; % counter ? tspnext = exprnd(1,1,ncells); % time of next spike (in rescaled time) rprev = zeros(1,ncells); % Integrated rescaled time up to current point % --------------- Run dynamics --------------------------------------- while jbin <= rlen iinxt = jbin:min(jbin+nbinsPerEval-1,rlen); % Bins to update in this iteration nii = length(iinxt); % Number of bins rrnxt = glmprs.nlfun(Itot(iinxt,:))*dt; % Cond Intensity rrcum = cumsum(rrnxt+[rprev;zeros(nii-1,ncells)],1); % Cumulative intensity if all(tspnext >= rrcum(end,:)) % No spike in this window jbin = iinxt(end)+1; rprev = rrcum(end,:); else % Spike! [ispks,jspks] = find(rrcum>=repmat(tspnext,nii,1)); spcells = unique(jspks(ispks == min(ispks))); % cell number(s) ispk = iinxt(min(ispks)); % time bin of spike(s) rprev = rrcum(min(ispks),:); % grab accumulated history to here % Record this spike mxi = min(rlen, ispk+hlen); % determine bins for adding h current iiPostSpk = ispk+1:mxi; for ic = 1:length(spcells) icell = spcells(ic); nsp(icell) = nsp(icell)+1; tsp{icell}(nsp(icell),1) = ispk*dt; if ~isempty(iiPostSpk) Itot(iiPostSpk,:) = Itot(iiPostSpk,:)+ih(1:mxi-ispk,:,icell); end rprev(icell) = 0; % reset this cell's integral tspnext(icell) = exprnd(1); % draw RV for next spike in this cell end jbin = ispk+1; % Move to next bin % -- Update # of samples per iter --- muISI = jbin/(sum(nsp)); nbinsPerEval = max(20, round(1.5*muISI)); end end % Remove any extra bins from tsp and compute binned spike train 'sps' sps = zeros(rlen,ncells); for jj = 1:ncells tsp{jj} = tsp{jj}(1:nsp(jj)); if ~isempty(tsp{jj}) sps(round(tsp{jj}/dt),jj) = 1; end end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/makeBasis_PostSpike.m
.m
3,133
90
function [iht, ihbas, ihbasis] = makeBasis_PostSpike(ihprs,dt) % [iht, ihbas, ihbasis] = makeBasis_PostSpike(ihprs,dt) % % Make nonlinearly stretched basis consisting of raised cosines % ------- % Inputs: % prs = param structure with fields: % ncols = # of basis vectors % hpeaks = 2-vector containg [1st_peak last_peak], the peak % location of first and last raised cosine basis vectors % b = offset for nonlinear stretching of x axis: y = log(x+b) % (larger b -> more nearly linear stretching) % absref = absolute refractory period (optional). If specified, % this param creates an additional "square" basis % vector with support n [0,absref] (i.e., with a hard % cutoff at absref) % % dt = grid of time points for representing basis % -------- % Outputs: iht = time lattice on which basis is defined % ihbas = orthogonalized basis % ihbasis = original raised cosine (non-orthogonal) basis % % ------------- % Example call: % % ihbasprs.ncols = 5; % ihbasprs.hpeaks = [.1 2]; % ihbasprs.b = .5; % ihbasprs.absref = .1; %% (optional) % [iht,ihbas,ihbasis] = makeBasis_PostSpike(ihprs,dt); ncols = ihprs.ncols; b = ihprs.b; hpeaks = ihprs.hpeaks; if isfield(ihprs, 'absref') absref = ihprs.absref; else absref = 0; end % Check input values if (hpeaks(1)+b) < 0 error('b + first peak location: must be greater than 0'); end if absref >= dt % use one fewer "cosine-shaped basis vector ncols = ncols-1; elseif absref > 0 warning('Refractory period too small for time-bin sizes'); end % nonlinearity for stretching x axis (and its inverse) nlin = @(x)log(x+1e-20); invnl = @(x)exp(x)-1e-20; % inverse nonlinearity ff = @(x,c,dc)(cos(max(-pi,min(pi,(x-c)*pi/dc/2)))+1)/2; % raised cosine basis vector % Generate basis of raised cosines if ncols > 1 yrnge = nlin(hpeaks+b); % nonlinearly transformed first & last bumps db = diff(yrnge)/(ncols-1); % spacing between cosine bump peaks ctrs = yrnge(1):db:yrnge(2); % centers (peak locations) for basis vectors else % if only 1 basis function if length(hpeaks)==1 hpeaks = [0 hpeaks]; % place imaginary 1st bump at 0 end yrnge = nlin(hpeaks+b); % nonlinearly transformed first & last bumps ncolseff = 3; % use 3 "effective" number of basis funcs, so left edge decays at 0 db = diff(yrnge)/(ncolseff-1); % spacing between cosine bump peaks ctrs = yrnge(2); % centers (peak locations) for basis vectors end % Make basis mxt = invnl(yrnge(2)+2*db)-b; % maximum time bin iht = (dt:dt:mxt)'; nt = length(iht); % number of points in iht ihbasis = ff(repmat(nlin(iht+b), 1, ncols), repmat(ctrs, nt, 1), db); % create first basis vector as step-function for absolute refractory period if absref >= dt ii = find(iht<absref); ih0 = zeros(size(ihbasis,1),1); ih0(ii) = 1; ihbasis(ii,:) = 0; ihbasis = [ih0,ihbasis]; end % compute orthogonalized basis ihbas = orth(ihbasis);
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/makeBasis_StimKernel.m
.m
2,344
61
function [kbasorth,kbasis] = makeBasis_StimKernel(kbasprs, nkt); % [kbasorth, kbasis] = makeBasis_StimKernel(kbasprs, nkt); % % Generates a basis consisting of raised cosines and several columns of % identity matrix vectors for temporal structure of stimulus kernel % % Args: kbasprs = struct with fields: % neye = number of identity basis vectors at front % ncos = # of vectors that are raised cosines % kpeaks = 2-vector, with peak position of 1st and last vector, % relative to start of cosine basis vectors (e.g. [0 10]) % b = offset for nonlinear scaling. larger values -> more linear % scaling of vectors. bk must be >= 0 % nkt = number of time samples in basis (optional) % % Output: % kbasorth = orthogonal basis % kbasis = standard raised cosine (non-orthongal) basis neye = kbasprs.neye; ncos = kbasprs.ncos; kpeaks = kbasprs.kpeaks; b = kbasprs.b; kdt = 1; % spacing of x axis must be in units of 1 % nonlinearity for stretching x axis (and its inverse) nlin = @(x)log(x+1e-20); invnl = @(x)exp(x)-1e-20; % inverse nonlinearity % Generate basis of raised cosines yrnge = nlin(kpeaks+b); db = diff(yrnge)/(ncos-1); % spacing between raised cosine peaks ctrs = yrnge(1):db:yrnge(2); % centers for basis vectors mxt = invnl(yrnge(2)+2*db)-b; % maximum time bin kt0 = (0:kdt:mxt)'; nt = length(kt0); % number of points in iht ff = @(x,c,dc)(cos(max(-pi,min(pi,(x-c)*pi/dc/2)))+1)/2; % raised cosine basis vector kbasis0 = ff(repmat(nlin(kt0+b), 1, ncos), repmat(ctrs, nt, 1), db); % Concatenate identity-vectors nkt0 = size(kt0,1); kbasis = [[eye(neye); zeros(nkt0,neye)] [zeros(neye, ncos); kbasis0]]; kbasis = flipud(kbasis); % flip so fine timescales are at the end. nkt0 = size(kbasis,1); if nargin > 1 if nkt0 < nkt %fprintf('>>>> makeBasis_StimKernel: Padding basis with %d rows of zeros\n', nkt-nkt0); kbasis = [zeros(nkt-nkt0,ncos+neye); kbasis]; elseif nkt0 > nkt %fprintf('>>>> makeBasis_StimKernel: Removing %d rows from basis\n', nkt0-nkt); kbasis = kbasis(end-nkt+1:end,:); end end kbasis = bsxfun(@rdivide,kbasis,sqrt(sum(kbasis.^2))); % normalize columns to be unit vectors kbasorth = orth(kbasis);
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/simpleSTC.m
.m
6,243
171
function [STA,STC,RawMu,RawCov] = simpleSTC(Stim,sp,nkt,maxsize) % [STA,STC,RawMu,RawCov] = simpleSTC(Stim,sp,nkt,maxsize) % % Computes mean and covariance of spike-triggered (or response weighted) % stimuli and raw stimuli % % INPUT: % Stim [N x M] - stimulus matrix; 1st dimension = time, 2nd dimension = space % sp [N x 1] - column vector of spike count in each time bin (can be sparse) % [Nsp x 1] - or, vector of spike times % nkt [1 x 1] - # time samples to consider to be part of the stimulus % MaxSize [1 x 1] (optional) - max # of floats to store while computing cov % (smaller = slower, but less memory required) % % OUTPUT: % STA [nkt x M] - spike-triggered average (reshaped as matrix) % STC [nkt*M x nkt*M] - spike-triggered covariance (covariance around the mean); % RawMu [nkt*M x 1] - mean of raw stimulus ensemble % RawCov [nkt*M x nkt*M] - covariance of raw stimulus ensemble % % Notes: % (1) Ignores response before nkt bins % (2) Faster if only 2 output arguments (raw mean and covariance not computed) % (3) Reduce 'maxsize' if getting "out of memory" errors % (4) If inputs are not spike times or counts, response-weighted covariance may be more appropriate. % See "simpleRWC.m". % % -------- % Details: % -------- % Let X = "valid" design matrix (from makeStimRows) % Y = sp (spike train) % nsp = sum(Y); % number of spikes (sum of responses) % N = size(Stim,1); % number of stimuli % % then STA = X'*Y / nsp % STC = X'*(X.*repmat(Y,1,ncols)))/(nsp-1) - STA*STA'*nsp/(nsp-1) % RawMu = sum(X)/N; % RawCov = X*X'/(N-1); % % Copyright 2010 Pillow Lab. All rights reserved. % $Id$ %-------- Parse inputs --------------------------- if nargin < 4 maxsize = 1e9; % max chunk size; decrease if getting "out of memory" end [slen,swid] = size(Stim); % stimulus size (# time bins x # spatial bins). nsp = length(sp); % Convert list of spike times to spike-count vector, if necessary if (nsp ~= slen) fprintf(1, 'simpleSTC: converting spike times to counts\n'); sp = hist(sp,1:slen)'; end sp(1:nkt-1) = 0; % Remove spikes before time n spsum = sum(sp); % Sum of spikes if (spsum == 0) spsum = 2; end nspnds = sum(sp~=0); % --------------------------------------------------- % 1. Compute only the spike-triggered STA and STC if nargout <= 2 Msz = nspnds*swid*nkt; % size of design matrix for spiking stimuli iisp = find(sp); splen = length(iisp); if Msz < maxsize % Compute in one chunk if small enough SS = makeStimRows(Stim, nkt,iisp); STA = (SS'*sp(iisp))/spsum; if nargout > 1 STC = SS'*bsxfun(@times,SS,sp(iisp))/(spsum-1) - STA*STA'*spsum/(spsum-1); end else % Compute in multiple chunks if too large nchunk = ceil(Msz/maxsize); chunksize = ceil(length(iisp)/nchunk); fprintf(1, 'simpleSTC: using %d chunks to compute STA/STC\n', nchunk); % Initialize on 1st chunk i0 = 1; imx = chunksize; SS = makeStimRows(Stim,nkt,iisp(i0:imx)); STA = (sp(iisp(i0:imx))'*SS)'; if nargout > 1 STC = SS'*bsxfun(@times,SS,sp(iisp(i0:imx))); end % Compute for remaining chunks for j = 2:nchunk i0 = chunksize*(j-1)+1; imx = min(chunksize*j, splen); SS = makeStimRows(Stim,nkt,iisp(i0:imx)); STA = STA + (sp(iisp(i0:imx))'*SS)'; if nargout > 1 STC = STC + SS'*bsxfun(@times,SS,sp(iisp(i0:imx))); end end % normalize by number of samples STA = STA/spsum; if nargout > 1 STC = STC/(spsum-1) - STA*STA'*spsum/(spsum-1); end end % ----------------------------------------------------------------------- % 2. Compute both the spike-triggered and raw stimulus means and covariances else sp = sp(nkt:end); % Ignore spikes before time bin nkt slen = length(sp); % length of time indices for stim and spikes Msz = slen*swid*nkt; % Size of full stimulus matrix if Msz < maxsize % Check if stimulus is small enough to do in one chunk SS = makeStimRows(Stim,nkt,'valid'); % Convert stimulus to matrix where each row is one stim % Compute raw mean and covariance RawMu = mean(SS)'; RawCov = (SS'*SS)/(slen-1)-RawMu*RawMu'*slen/(slen-1); % Compute spike-triggered mean and covariance iisp = find(sp>0); spvec = sp(iisp); STA = (spvec'*SS(iisp,:))'/spsum; STC = SS(iisp,:)'*bsxfun(@times,SS(iisp,:),spvec)/(spsum-1) ... - STA*(STA'*spsum/(spsum-1)); else % Compute Full Stim matrix in chunks, compute mean and cov on chunks nchunk = ceil(Msz/maxsize); chunksize = ceil(slen/nchunk); fprintf(1, 'simpleSTC: using %d chunks to compute covariance\n', nchunk); % Compute statistics on first chunk SS = makeStimRows(Stim(1:chunksize+nkt-1,:),nkt,'valid'); % convert stimulus to "full" version spvec = sp(1:chunksize); iisp = find(spvec>0); RawMu = sum(SS)'; RawCov = SS'*SS; STA = (spvec(iisp)'*SS(iisp,:))'; STC = SS(iisp,:)'*bsxfun(@times,SS(iisp,:),spvec(iisp)); % add to mean and covariance for remaining chunks for j = 2:nchunk; i0 = chunksize*(j-1)+1; % starting index for chunk imax = min(slen,chunksize*j); % ending index for chunk SS = makeStimRows(Stim(i0:imax+nkt-1,:),nkt,'valid'); spvec = sp(i0:imax); iisp = find(spvec); RawMu = RawMu + sum(SS)'; RawCov = RawCov +SS'*SS; STA = STA + (spvec(iisp)'*SS(iisp,:))'; STC = STC + SS(iisp,:)'*bsxfun(@times,SS(iisp,:),spvec(iisp)); end % divide means and covariances by number of samples RawMu = RawMu/slen; RawCov = RawCov/(slen-1) - RawMu*RawMu'*slen/(slen-1); STA = STA/spsum; STC = STC/(spsum-1) - STA*STA'*spsum/(spsum-1); end end STA = reshape(STA,[],swid); % reshape to have same width as stimulus
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/makeSimStruct_GLM.m
.m
3,025
70
function S = makeSimStruct_GLM(nkt,dtStim,dtSp) % S = makeSimStruct_GLM(nkt,dtStim,dtSp); % % Creates a structure with default parameters for a GLM model % % Input: nkt = number of time bins for stimulus filter % dtStim = bin size for sampling of stimulus kernel ('ih'), in s % dtSp = bin size for sampling post-spike kernel ('k'), in s % (should evenly divde dtStim) % % Struct fields (model params): % 'filt' - stimulus filter % 'nlfun' - nonlinearity (exponential by default) % 'dc' - dc input to cell % 'ih' - post-spike current % 'ihbas' - basis for post-spike current % 'iht' - time lattice for post-spike current % 'dtsim' - default time bin size for simulation % 'ihbasprs' - basis for post-spike current % Check that stimulus bins bigger than spike bins assert (dtStim>=dtSp, 'dtStim must be >= dtSp'); % Check that spike bin size evenly divides stimulus bin size if mod(dtStim,dtSp)~=0 warning('makeSimStruct_GLM: dtSp doesn''t evenly divide dtStim: rounding dtSp to nearest even divisor'); dtSp = dtStim/round(dtStim/dtSp); fprintf('dtSp reset to: %.3f\n', dtSp); end % Create a default (temporal) stimulus filter tk = (0:nkt-1)'; % time indices for made-up filter b1 = nkt/32; b2 = nkt/16; k1 = 1/(gamma(6)*b1)*(tk/b1).^5 .* exp(-tk/b1); % Gamma pdfn k2 = 1/(gamma(6)*b2)*(tk/b2).^5 .* exp(-tk/b2); % Gamma pdf k = flipud(k1-k2./1.5+1e-5); k = k./norm(k)/2; % % ----- Represent this filter (approximately) in temporal basis ------- ktbasprs.neye = min(5,nkt); % Number of "identity" basis vectors near time of spike; ktbasprs.ncos = min(5,nkt); % Number of raised-cosine vectors to use ktbasprs.kpeaks = [0 ((nkt-ktbasprs.neye)/2)]; % Position of 1st and last bump ktbasprs.b = 1; % Offset for nonlinear scaling (larger -> more linear) ktbas = makeBasis_StimKernel(ktbasprs,nkt); k = ktbas*(ktbas\k); % -- Nonlinearity ------- nlinF = @expfun; % Other natural choice: nlinF = @(x)log(1+exp(x)); % --- Make basis for post-spike (h) filter ------ ihbasprs.ncols = 5; % Number of basis vectors for post-spike kernel ihbasprs.hpeaks = [dtSp dtSp*50]; % Peak location for first and last vectors ihbasprs.b = dtSp*5; % Determines how nonlinear to make spacings ihbasprs.absref = []; % absolute refractory period (optional) [iht,~,ihbasis] = makeBasis_PostSpike(ihbasprs,dtSp); ih = ihbasis*[-10 .5 1 .25 -1]'; % h current % Place parameters in structure S = struct(... 'k', 2*k, ... % stimulus filter 'nlfun', nlinF, ... % nonlinearity 'dc', 1.5, ... % dc input (constant) 'ih', ih, ... % post-spike current 'iht', iht, ... % time indices of aftercurrent 'dtStim', dtStim,... % time bin size for stimulus (s) 'dtSp', dtSp, ... % time bin size for spikes (s) 'ihbasprs', ihbasprs, ... % params for ih basis 'ktbasprs', ktbasprs); % params for k time-basis
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/sameconv.m
.m
1,679
58
function Y = sameconv(X, F) % SAMECONV - causally filters X with F and returns output of same height % % Y = sameconv(X, F) % % F is not "flipped", so first element of Y is last row of F dot product with % first row of X. % % Inputs: % X [NxM] - tall matrix % F [RxM] - short matrix (same width as X) % % Output: % Y [Nx1] - X filtered causally with F. % (first element is F(end,:)*X(1,:)' ) % % Attempts to use fastest method based on length and width of F. % (i.e., using either matrix arithmetic or matlab's "conv2"). % % ex) sameconv(+- 0 0 -+, +- 1 4 -+) --> +- 0 -+ % | 1 0 | | 2 5 | | 3 | % | 0 0 | +- 3 6 -+ | 2 | % | 0 0 | | 1 | % | 0 0 | | 0 | % | 0 1 | | 6 | % +- 0 0 -+ +- 5 -+ % % (updated: 14/03/2011 JW Pillow) % % Copyright 2011 Pillow Lab. % $Id$ [nx, xwid] = size(X); [nf, fwid] = size(F); assert(xwid == fwid, 'X and F must have same second dimension'); % Decide which method to use based on size of F if log2(nf)+3>2*log2(fwid) % Use CONV2 if F is tall and skinny % conv2 doesn't take sparse matrix if issparse(X); X = full(X); end if issparse(F); F = full(F); end Y = conv2([zeros(nf-1,xwid); X], rot90(F,2), 'valid'); else % Use Matrix Operations if F is short and fat. ny = nx+nf-1; % # of columns in full convolution Y = zeros(ny,1); yy = X*F'; for j = 1:nf Y(j:j+nx-1) = Y(j:j+nx-1) + yy(:,nf-j+1); end % Extract only the first part Y = Y(1:nx,:); end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_misc/makeSimStruct_GLMcpl.m
.m
2,094
89
function S = makeSimStruct_GLMcpl(varargin); % S = makeSimStruct_GLMcpl(gg1, gg2, gg3, ...); % % Creates a structure for simulating coupled GLM from multiple % single-neuron GLMs. % % Input: structures params of glms to combine ncells = nargin; gg = varargin{1}; [klen,kwid] = size(gg.k); kk = zeros(klen,kwid,ncells); % Stimulus filters nh = length(gg.iht); ih = zeros(nh,ncells,ncells); % post-spike & coupling kernels ihflag = 0; % Check for spike-history kernel if ~isempty(gg.ih), ihflag = 1; end ih2flag = 0; % Check for presence of coupling kernels if size(gg.ih,2)>1 ih2flag = 1; end if isfield(gg, 'ih2'); if ~isempty(gg.ih2); ih2flag = 1; end; end ihbasflag = 0; % Basis for history kernel? if isfield(gg, 'ihbas'); if ~isempty(gg.ihbas) ihbasflag = 1; end; end ihbas2flag=0; % Distinct basis for coupling kernels? if isfield(gg, 'ihbas2'); if ~isempty(gg.ihbas2) ihbas2flag = 1; end; end % Add cells to single param structure nltestvals = [0, 1, 2]; for j = 1:ncells gg = varargin{j}; kk(:,:,j) = gg.k; dc(j) = gg.dc; if isfield(gg, 'nlfun') nlfuns{j} = gg.nlfun; else nlfuns{j} = @exp; end NLtst(j,:) = nlfuns{j}(nltestvals); if ihflag % Post-spike kernel if ihbasflag ih(:,j,j) = gg.ihbas*gg.ih(:,1); else ih(:,j,j) = gg.ih(:,1); end end if ih2flag % Coupling kernels if ihbas2flag ih(:,setdiff(1:ncells,j),j) = gg.ihbas2*gg.ih2; elseif ihbasflag ih(:,setdiff(1:ncells,j),j) = gg.ihbas*gg.ih(:,2:end); else ih(:,setdiff(1:ncells,j),j) = gg.ih(:,2:end); end end end S.k = kk; S.dc = dc; S.ih = ih; S.iht = varargin{1}.iht; S.dtStim = varargin{1}.dtStim; S.dtSp = varargin{1}.dtSp; if all(min(NLtst)==max(NLtst)) S.nlfun = nlfuns{1}; else S.nlfun = @(x)multiNLfun(x,nlfuns); end if isfield(gg,'ihbasprs') S.ihbasprs = gg.ihbasprs; end if isfield(gg,'ihbasprs2') S.ihbasprs2 = gg.ihbasprs2; end
MATLAB
2D
pillowlab/GLMspiketools
demos/demo5_GLM_10CoupledNeurons.m
.m
8,057
205
% testscript3_GLM_coupled.m % % Demo script for simulating and fitting a coupled GLM (2 neurons). % % Notes: % - Fitting code uses same functions as for single-cell responses. % - Simulation code requires new structures / functions % (due to the need to pass activity between neurons) % Make sure paths are set (assumes this script called from 'demos' directory) cd ..; setpaths_GLMspiketools; cd demos/ %% ===== 1. Set parameters for simulating a GLM ============ % dtStim = .001; % Bin size for simulating model & computing likelihood (in units of stimulus frames) dtSp = .001; % Bin size for simulating model & computing likelihood (must evenly divide dtStim); nkt = 1; % Number of time bins in stimulus filter k %% FIX THIS SO IT DOESN'T MAKE NAN gg = makeSimStruct_GLM(nkt,dtStim,dtSp); % Create GLM struct with default params %% Set up basis for coupling filters % Make basis for self-coupling term ihbasprs.ncols = 3; % number of basis vectors ihbasprs.hpeaks = [.001, .005]; % peak of 1st and last basis vector ihbasprs.b = .001; % scaling (smaller -> more logarithmic scaling) ihbasprs.absref = []; % absolute refractory period basis vector (optional) % Make basis [iht,ihbas,ihbasis] = makeBasis_PostSpike(ihbasprs,dtSp); nht = length(iht); % number of bins % Make basis for cross-coupling term ihbasprs2.ncols = 1; % number of basis vectors ihbasprs2.hpeaks = [0.001,.005]; % put peak at 5ms and "effective" 1st peak at 0 ihbasprs2.b = .001; % smaller -> more logarithmic scaling ihbasprs2.absref = []; % no abs-refracotry period for this one % Make basis [iht2,ihbas2,ihbasis2] = makeBasis_PostSpike(ihbasprs2,dtSp); nht2 = length(iht2); % pad to put them into the same time bins if nht2>nht % padd ih1 with zeros iht = iht2; zz = zeros(nht2-nht,ihbasprs.ncols); ihbas = [ihbas;zz]; ihbasis = [ihbasis;zz]; nht=nht2; elseif nht2<nht % padd ih1 with zeros iht2 = iht; zz = zeros(nht-nht2,ihbasprs2.ncols); ihbas2 = [ihbas2;zz]; ihbasis2 = [ihbasis2;zz]; nht2=nht; end % plot it subplot(211); plot(iht, ihbasis, ihbasprs.hpeaks, 1, '*', 'linewidth', 2); xlabel('time after spike (ms)'); title('post-spike basis'); subplot(212); plot(iht2, ihbasis2, ihbasprs2.hpeaks, 1, '*', 'linewidth', 2); xlabel('time after spike (ms)'); title('coupling basis'); %% Set self-coupling filter shapes wself = [-10; 3.2; -1]; % weights for self-coupling term ihself = ihbasis*wself; % self-coupling filter wcpl = 0.5; % weights for cross-coupling term ihcpl = ihbasis2*wcpl; % cross-coupling filter clf; plot(iht, exp(ihself), iht, exp(ihcpl), iht, iht*0+1, 'k--'); legend('self-coupling', 'cross-coupling'); xlabel('time lag (s)'); ylabel('gain (sp/s)'); %% Set up multi-neuron GLM nneur = 10; k = randn(10,1)*0.5; % stimulus weights gg.k = permute(k,[2,3,1]); % stimulus weights gg.dc = 2+(rand(1,nneur)-0.5); % Generate random coupling strengths WW = randn(nneur); % coupling strengths WW = WW-diag(diag(WW)-1); % set diagonal to zero (for self-coupling) gg.iht = iht; gg.ih = zeros(nht,nneur,nneur); % Insert coupling filters for jj = 1:nneur gg.ih(:,:,jj) = ihcpl*WW(jj,:); % cross-coupling input weights to neuron jj gg.ih(:,jj,jj) = ihself*WW(jj,jj); % self-coupling weights end % Plot filters lw = 2; % linewidth subplot(131); plot(gg.iht, gg.ih(:,:,1),gg.iht,gg.iht*0, 'k--', 'linewidth', lw); title('incoming filters: cell 1'); subplot(132); plot(gg.iht, gg.ih(:,:,2),gg.iht,gg.iht*0, 'k--', 'linewidth', lw); title('incoming filters: cell 2'); subplot(133); plot(gg.iht, gg.ih(:,:,3),gg.iht,gg.iht*0, 'k--', 'linewidth', lw); %% ===== 2. Run short simulation for visualization purposes ========= % slen = 200; % Stimulus length (frames) & width (# pixels) swid = 1; % width of stimulus stimsd = 1; % contrast of stimulus Stim = stimsd*randn(slen,swid); % Gaussian white noise stimulus [tsp,~,Itot,Istm] = simGLM(gg, Stim); % Simulate GLM response Isp = Itot-Istm; % net spike-history output % ==== Plot some traces of simulated response ======== tt = (dtSp:dtSp:slen*dtStim)'; subplot(131) % % ==== neuron 1 =========== plot(tt, Istm(:,1), 'k', tt, Isp(:,1), 'r', tt, tt*0, 'k--', ... tsp{1}, ones(size(tsp{1})), 'ro'); title('cell 1'); axis tight; xlabel('time (s)'); ylabel('filter output'); legend('stim filter + dc', 'spike-hist filter'); subplot(132) % ==== neuron 2 =========== plot(tt, Istm(:,2), 'k', tt, Isp(:,2), 'r', tt, tt*0, 'k--', ... tsp{2}, ones(size(tsp{2})), 'ro'); title('cell 2'); axis tight; xlabel('time (s)'); subplot(133) % ==== neuron 3 =========== plot(tt, Istm(:,3), 'k', tt, Isp(:,3), 'r', tt, tt*0, 'k--', ... tsp{3}, ones(size(tsp{3})), 'ro'); title('cell 3'); axis tight; xlabel('time (s)'); %% ===== 3. Generate some training data =============================== %% slen = 5e5; % Stimulus length (in bins); Stim = stimsd*randn(slen,swid); % Gaussian white noise stimulus [tsp,sps,Itot,ispk] = simGLM(gg,Stim); % run model %% ===== 4. Do ML fitting for all neurons in a loop =================== %% % Initialize param struct for fitting ggInit = makeFittingStruct_GLM(dtStim,dtSp); % Initialize params for fitting struct % Initialize fields (using h bases computed above) ggInit.ktbas = 1; % k basis ggInit.ihbas = ihbas; % h self-coupling basis ggInit.ihbas2 = ihbas2; % h coupling-filter basis nktbasis = 1; % number of basis vectors in k basis nhbasis = size(ihbas,2); % number of basis vectors in h basis nhbasis2 = size(ihbas2,nneur-1); % number of basis vectors in h basis ggInit.kt = 1; % initial params from scaled-down sta ggInit.k = 1; % initial setting of k filter ggInit.ihw = zeros(nhbasis,1); % init params for self-coupling filter ggInit.ihw2 = zeros(nhbasis2,nneur-1); % init params for cross-coupling filter ggInit.ih = [ggInit.ihbas*ggInit.ihw ggInit.ihbas2*ggInit.ihw2]; ggInit.iht = iht; ggInit.dc = 0; % Initialize dc term to zero ggInit.sps = sps(:,1); % spikes from 1 cell ggInit.sps2 = sps(:,2:nneur); % spikes from all ggInit.couplednums = 2:nneur; % cell numbers of cells coupled to this one ggfit(1:nneur) = ggInit; % initialize fitting struct kfit = zeros(nneur,1); dcfit = zeros(nneur,1); for jj = 1:nneur couplednums = setdiff(1:nneur,jj); % cell numbers of cells coupled to this one % Set spike responses for cell 1 and coupled cell ggInit.sps = sps(:,jj); ggInit.sps2 = sps(:,couplednums); ggInit.couplednums = couplednums; % number of cell coupled to this one (for clarity) % Do ML fitting fprintf('==== Fitting filters to neuron %d ==== \n', jj) opts = {'display', 'iter', 'maxiter', 100}; ggfit(jj) = MLfit_GLM(ggInit,Stim,opts); % do ML fitting kfit(jj) = ggfit(jj).k; dcfit(jj) = ggfit(jj).dc; end %% ===== 5. Plot true and fitted params for first few cells ============================================= %% clf reset; colors = get(gca,'colororder'); ncolrs = size(colors,1); % number of traces to plot for each cell ymax = max(exp([ggfit(1).ih(:);ggfit(2).ih(:);ggfit(3).ih(:);gg.ih(:)])); % max of y range for jj = 1:10 subplot(2,5,jj); % --Spike filters cell 1 % ------------- ccpl = setdiff(1:nneur,jj); % coupled cells cnums = [jj, ccpl]; plot(gg.iht, exp(gg.ih(:,cnums(1:ncolrs),jj)), ggfit(jj).iht, exp(ggfit(jj).ih(:,1:ncolrs)), '--', 'linewidth', lw); hold on; plot(gg.iht, gg.iht*0+1, 'k'); hold off; title(sprintf('cell %d filters',jj)); axis tight; set(gca,'ylim',[0,ymax]); if jj==1 ylabel('gain (sp/s)'); xlabel('time after spike (s)'); end end % Print true and recovered params: fprintf('\n---------------------------------\n'); fprintf('k: true est | dc: true est\n'); fprintf('---------------------------------\n'); for jj = 1:nneur fprintf(' %5.2f %5.2f %5.2f %5.2f\n', [gg.k(jj), kfit(jj), gg.dc(jj), dcfit(jj)]); end
MATLAB
2D
pillowlab/GLMspiketools
demos/demo2_GLM_spatialStim.m
.m
6,074
157
% demo2_GLM_spatialStim.m % % Test code for simulating and fitting the GLM with a 2D stimulus filter % (time x 1D space), with both traditional and bilinear parametrization % of the stimulus kernel. % Make sure paths are set (assumes this script called from 'demos' directory) cd ..; setpaths_GLMspiketools; cd demos/ %% 1. Set parameters and display for GLM ============= % dtStim = .01; % Bin size for stimulus (in seconds). (Equiv to 100Hz frame rate) dtSp = .001; % Bin size for simulating model & computing likelihood (must evenly divide dtStim); % Make a temporal filter nkt = 20; % Number of time bins in stimulus filter k kt = (normpdf(1:nkt,3*nkt/4,1.5)-.5*normpdf(1:nkt,nkt/2,3))'; % Make a spatial filter; nkx = 10; xxk = (1:1:nkx)'; % pixel locations ttk = dtStim*(-nkt+1:0)'; % time bins for filter kx = 1./sqrt(2*pi*4).*exp(-(xxk-nkx/2).^2/5); k = kt*kx'; % Make space-time separable filter k = k./norm(k(:))/1.5; % Insert into glm structure (created with default history filter) ggsim = makeSimStruct_GLM(nkt,dtStim,dtSp); % Create GLM structure with default params ggsim.k = k; % Insert into simulation struct ggsim.dc = 2; % === Make Fig: model params ======================= clf; subplot(3,3,[1 4]); % ------------------------------------------ plot(kt,ttk); axis tight; set(gca, 'ydir', 'reverse'); ylabel('time (frames)'); subplot(3,3,[2 3 5 6]); % -------------------------------------- imagesc(xxk,ttk,ggsim.k); colormap gray; axis off; title('stimulus kernel k'); subplot(3,3,8:9); % ---------------------------------------- plot(xxk,kx); axis tight; set(gca, 'xlim', [.5 nkx+.5]); xlabel('space (pixels)'); %% 2. Generate some training data ======================================== slen = 10000; % Stimulus length (frames); More samples gives better fit Stim = round(rand(slen,nkx))*2-1; % Run model on long, binary stimulus [tsp,sps,Itot,Istm] = simGLM(ggsim,Stim); % run model nsp = length(tsp); % --- Make plot of first 0.5 seconds of data -------- tlen = 0.5; ttstim = dtStim:dtStim:tlen; iistim = 1:length(ttstim); ttspk = dtSp:dtSp:tlen; iispk = 1:length(ttspk); spinds = sps(iispk)>0; subplot(311); imagesc([0 tlen], [1 nkx], Stim(iistim,:)'); title('stimulus'); ylabel('pixel'); subplot(312); semilogy(ttspk,exp(Itot(iispk)),ttspk(spinds), exp(Itot(spinds)), 'ko'); ylabel('spike rate (sp/s)'); title('conditional intensity (and spikes)'); subplot(313); Isp = Itot-Istm; % total spike-history filter output plot(ttspk,Istm(iispk), ttspk,Isp(iispk)); axis tight; legend('k output', 'h output'); xlabel('time (s)'); ylabel('log intensity'); title('filter outputs'); %% 3. Fit GLM (traditional version) via max likelihood % Compute the STA sps_coarse = sum(reshape(sps,[],slen),1)'; % bin spikes in bins the size of stimulus sta0 = simpleSTC(Stim,sps_coarse,nkt); % Compute STA sta = reshape(sta0,nkt,[]); % reshape it to match dimensions of true filter exptmask= []; % Not currently supported! % Initialize params for fitting -------------- % Set params for fitting, including bases nkbasis = 8; % number of basis vectors for representing k nhbasis = 8; % number of basis vectors for representing h hpeakFinal = .1; % time of peak of last basis vector for h gg0 = makeFittingStruct_GLM(dtStim,dtSp,nkt,nkbasis,sta,nhbasis,hpeakFinal); gg0.sps = sps; % Insert binned spike train into fitting struct gg0.mask = exptmask; % insert mask (optional) gg0.ihw = randn(size(gg0.ihw))*1; % initialize spike-history weights randomly % Compute conditional intensity at initial parameters [negloglival0,rr] = neglogli_GLM(gg0,Stim); fprintf('Initial negative log-likelihood: %.5f\n', negloglival0); % Do ML estimation of model params opts = {'display', 'iter', 'maxiter', 100}; [gg1, negloglival1a] = MLfit_GLM(gg0,Stim,opts); % do ML (requires optimization toolbox) %% 4. Fit GLM ("bilinear stim filter version") via max likelihood % Initialize params for fitting -------------- k_rank = 1; % Number of column/row vector pairs to use gg0b = makeFittingStruct_GLMbi(k_rank,dtStim,dtSp,nkt,nkbasis,sta,nhbasis,hpeakFinal); gg0b.sps = sps; gg0b.mask = exptmask; logli0b = neglogli_GLM(gg0b,Stim); % Compute logli of initial params fprintf('Initial value of negative log-li (GLMbi): %.3f\n', logli0b); % Do ML estimation of model params opts = {'display', 'iter'}; [gg2, negloglival2] = MLfit_GLMbi(gg0b,Stim,opts); % do ML (requires optimization toolbox) %% 5. Plot results ==================== %figure(3); subplot(231); % True filter % --------------- imagesc(ggsim.k); colormap gray; title('True Filter');ylabel('time'); subplot(232); % sta % ------------------------ imagesc(sta); title('raw STA'); ylabel('time'); subplot(233); % sta-projection % --------------- imagesc(gg0.k); title('low-rank STA'); subplot(234); % estimated filter % --------------- imagesc(gg1.k); title('ML estimate: full filter'); xlabel('space'); ylabel('time'); subplot(235); % estimated filter % --------------- imagesc(gg2.k); title('ML estimate: bilinear filter'); xlabel('space'); subplot(236); % ---------------------------------- plot(ggsim.iht,exp(ggsim.ih),'k', gg1.iht,exp(gg1.ihbas*gg1.ihw),'b',... gg2.iht, exp(gg2.ihbas*gg2.ihw), 'r'); axis tight; title('post-spike kernel'); xlabel('time after spike (s)'); legend('true','GLM','bilinear GLM'); % Errors in STA and ML estimate ktmu = normalizecols([mean(ggsim.k,2),mean(gg1.k,2),mean(gg2.k,2)]); kxmu = normalizecols([mean(ggsim.k)',mean(gg1.k)',mean(gg2.k)']); msefun = @(k)(sum((k(:)-ggsim.k(:)).^2)); % error function fprintf(['K-filter errors (GLM vs. GLMbilinear):\n', ... 'Temporal error: %.3f %.3f\n', ... ' Spatial error: %.3f %.3f\n', ... ' Total error: %.3f %.3f\n'], ... subspace(ktmu(:,1),ktmu(:,2)), subspace(ktmu(:,1),ktmu(:,3)), ... subspace(kxmu(:,1),kxmu(:,2)), subspace(kxmu(:,1),kxmu(:,3)), ... msefun(gg1.k), msefun(gg2.k));
MATLAB
2D
pillowlab/GLMspiketools
demos/demo1_GLM_temporalStim.m
.m
4,719
116
% demo1_GLM_temporalStim.m % % Demo script for simulating and fitting a single-neuron GLM with 1D % (temporal) filter with exponential nonlinearity % Make sure paths are set (assumes this script called from 'demos' directory) cd ..; setpaths_GLMspiketools; cd demos/ %% 1. Set parameters and display for GLM % ============================= dtStim = .01; % Bin size for stimulus (in seconds). (Equiv to 100Hz frame rate) dtSp = .001; % Bin size for simulating model & computing likelihood (must evenly divide dtStim); nkt = 30; % Number of time bins in stimulus filter k ggsim = makeSimStruct_GLM(nkt,dtStim,dtSp); % Create GLM structure with default params % === Plot true model params ======================= clf; ttk = dtStim*(-nkt+1:0)'; % time relative to spike of stim filter taps subplot(221);plot(ttk, ggsim.k); title('stimulus kernel'); xlabel('time (s)'); subplot(222); % -------- plot(ggsim.iht, ggsim.iht*0, 'k--', ggsim.iht, ggsim.ih); title('post-spike kernel h'); axis tight; xlabel('time after spike (s)'); set(gca, 'ylim',[min(ggsim.ih)*1.1 max(ggsim.ih)*1.5]); subplot(224); % -------- [iht,ihbasOrthog,ihbasis] = makeBasis_PostSpike(ggsim.ihbasprs,dtSp); plot(ggsim.iht, ihbasis); title('basis for h'); axis tight; %% 2. Generate some training data %======================================== slen = 50000; % Stimulus length (frames); more samples gives better fit swid = 1; % Stimulus width (pixels); must match # pixels in stim filter % Make stimulus Stim = rand(slen,swid)*2-1; % Stimulate model to long, unif-random stimulus % Simulate model [tsp,sps,Itot,Istm] = simGLM(ggsim,Stim); % run model % --- Make plot of first 0.5 seconds of data -------- tlen = 0.5; ttstim = dtStim:dtStim:tlen; iistim = 1:length(ttstim); subplot(311); plot(ttstim,Stim(iistim)); title('stimulus'); subplot(312); ttspk = dtSp:dtSp:tlen; iispk = 1:length(ttspk); spinds = sps(iispk)>0; semilogy(ttspk,exp(Itot(iispk)),ttspk(spinds), exp(Itot(spinds)), 'ko'); ylabel('spike rate (sp/s)'); title('conditional intensity (and spikes)'); subplot(313); Isp = Itot-Istm; % total spike-history filter output plot(ttspk,Istm(iispk), ttspk,Isp(iispk)); axis tight; legend('k output', 'h output'); xlabel('time (s)'); ylabel('log intensity'); title('filter outputs'); %% 3. Setup fitting params %=================================================== % Compute the STA sps_coarse = sum(reshape(sps,[],slen),1)'; % bin spikes in bins the size of stimulus sta = simpleSTC(Stim,sps_coarse,nkt); % Compute STA sta = reshape(sta,nkt,[]); % reshape it to match dimensions of true filter % Set mask (if desired) exptmask= []; %[1 slen*dtStim]; % data range to use for fitting (in s). % Set params for fitting, including bases nkbasis = 8; % number of basis vectors for representing k nhbasis = 8; % number of basis vectors for representing h hpeakFinal = .1; % time of peak of last basis vector for h gg0 = makeFittingStruct_GLM(dtStim,dtSp,nkt,nkbasis,sta,nhbasis,hpeakFinal); gg0.sps = sps; % Insert binned spike train into fitting struct gg0.mask = exptmask; % insert mask (optional) gg0.ihw = randn(size(gg0.ihw))*1; % initialize spike-history weights randomly % Compute conditional intensity at initial parameters [negloglival0,rr] = neglogli_GLM(gg0,Stim); fprintf('Initial negative log-likelihood: %.5f\n', negloglival0); %% 4. Do ML fitting %===================== opts = {'display', 'iter', 'maxiter', 100}; % options for fminunc [gg1, negloglival] = MLfit_GLM(gg0,Stim,opts); % do ML (requires optimization toolbox) %% 5. Plot results ====================================================== ttk = -nkt+1:0; % time bins for stimulus filter subplot(221); % True filter plot(ttk, ggsim.k, 'k', ttk, sta./norm(sta)*norm(ggsim.k), ttk, gg1.k, 'r'); title('Stim filters'); legend('k_{true}', 'k_{STA}', 'k_{ML}', 'location', 'northwest'); % ---------------------------------- subplot(222); plot(ggsim.iht, ggsim.ih, gg1.iht, gg1.ih,'r', ggsim.iht, ggsim.iht*0, 'k--'); title('post-spike kernel'); axis tight; legend('h_{true}', 'h_{ML}', 'location', 'southeast'); % ---------------------------------- subplot(224); plot(ggsim.iht, exp(ggsim.ih), gg1.iht, exp(gg1.ih),'r', ggsim.iht,ggsim.iht*0+1,'k--'); title('exponentiated post-spike kernels'); xlabel('time since spike (s)'); ylabel('gain'); axis tight; legend('h_{true}', 'h_{ML}', 'location', 'southeast'); % Errors in STA and ML estimate (subspace angle between true k and estimate) fprintf('Filter estimation error (in radians)\n sta: %.3f\n ML: %.3f\n', ... subspace(ggsim.k,sta), subspace(ggsim.k,gg1.k));
MATLAB
2D
pillowlab/GLMspiketools
demos/demo3_GLM_coupled.m
.m
7,432
174
% testscript3_GLM_coupled.m % % Demo script for simulating and fitting a coupled GLM (2 neurons). % % Notes: % - Fitting code uses same functions as for single-cell responses. % - Simulation code requires new structures / functions % (due to the need to pass activity between neurons) % Make sure paths are set (assumes this script called from 'demos' directory) cd ..; setpaths_GLMspiketools; cd demos/ %% ===== 1. Set parameters and display for GLM ============ % dtStim = .01; % Bin size for simulating model & computing likelihood (in units of stimulus frames) dtSp = .001; % Bin size for simulating model & computing likelihood (must evenly divide dtStim); nkt = 20; % Number of time bins in stimulus filter k ggsim1 = makeSimStruct_GLM(nkt,dtStim,dtSp); % Create GLM struct with default params ggsim2 = makeSimStruct_GLM(nkt,dtStim,dtSp); % Create GLM struct with default params % Change second neuron's stimulus filter [ktbas,ktbasis] = makeBasis_StimKernel(ggsim2.ktbasprs,nkt); ggsim2.k = ktbasis*[0 0 .1 .25 .5 .5 .25 -.25 -.5 -.25]'; % more delayed filter ggsim = makeSimStruct_GLMcpl(ggsim1,ggsim2); % Make some coupling kernels [iht,ihbas,ihbasis] = makeBasis_PostSpike(ggsim.ihbasprs,dtSp); hhcpl = ihbasis*[.25;.25;.125;0;0]; hhcpl(:,2) = ihbasis*[-2;-1;0;.25;.25]; ggsim.ih(:,2,1) = hhcpl(:,2); % 2nd cell coupling to first ggsim.ih(:,1,2) = hhcpl(:,1); % 1st cell coupling to second % === Make Fig: model params ======================= ttk = (-nkt+1:0)'*dtStim; subplot(2,2,1); % ------------------------------- plot(ttk, ggsim.k(:,:,1)); title('neuron 1 stim kernel'); subplot(2,2,3); % ------------------------------- plot(ttk, ggsim.k(:,:,2)); title('neuron 2 stim kernel'); xlabel('time (frames)'); subplot(2,2,2); % -------------------------------- plot(ggsim.iht, exp(ggsim.ih(:,:,1)), ggsim.iht, ggsim.iht*0+1, 'k--'); title('spike kernels into Cell 1 (exponentiated)'); legend('from 1', 'from 2', 'location', 'northeast'); ylabel('gain'); subplot(2,2,4); % --------------------------------- plot(ggsim.iht, exp(ggsim.ih(:,:,2)), ggsim.iht, ggsim.iht*0+1, 'k--'); title('spike kernels into Cell 2 (exponentiated)'); legend('from 1', 'from 2', 'location', 'northeast'); ylabel('gain'); xlabel('time after spike (frames)'); %% ===== 2. Run short simulation for visualization purposes ========= % % slen = 50; % Stimulus length (frames) & width (# pixels) swid = size(ggsim.k,2); % stimulus width Stim = 2*randn(slen,swid); % Gaussian white noise stimulus [tsp,~,Itot,Istm] = simGLM(ggsim, Stim); % Simulate GLM response Isp = Itot-Istm; % net spike-history output % ==== Plot some traces of simulated response ======== tt = (dtSp:dtSp:slen*dtStim)'; subplot(321); %------------------------ plot(1:slen, Stim, 'k', 'linewidth', 2); title('GWN stimulus'); axis tight; subplot(3,2,3); %------------------------ plot(tt, Itot(:,1),'b',tsp{1}, max(Itot(:,1))*ones(size(tsp{1})), 'ro'); title('cell 1: net voltage + spikes'); axis([0 slen*dtStim, min(Itot(:,1)) max(Itot(:,1))*1.01]); ylabel('filter output'); subplot(3,2,4); %------------------------ plot(tt, Itot(:,2), 'b', tsp{2}, max(Itot(:,2))*ones(size(tsp{2})), 'ro'); title('cell 2: net voltage + spikes'); axis([0 slen*dtStim, min(Itot(:,2)) max(Itot(:,2))*1.01]); subplot(325) % -------------------------- plot(tt, Istm(:,1), 'k', tt, Isp(:,1), 'r'); title('stim-induced & spike-induced currents'); axis tight; xlabel('time (frames)'); ylabel('filter output'); subplot(326); %--------------------------- plot(tt, Istm(:,2), 'k', tt, Isp(:,2), 'r'); title('stim-induced & spike-induced currents'); axis tight; xlabel('time (frames)'); %% ===== 3. Generate some training data =============================== %% slen = 20000; % Stimulus length (frames); More samples gives better fit Stim = round(rand(slen,swid))-.5; % Run model on long, binary stimulus [tsp,sps,Itot,ispk] = simGLM(ggsim,Stim); % run model %% ===== 4. Fit cell #1 (with coupling from cell #2) =================== %% % Compute Spike-triggered averages sps1 = sum(reshape(sps(:,1),[],slen),1)'; % bin spikes in bins the size of stimulus sta1 = simpleSTC(Stim,sps1,nkt); % Compute STA 1 sta1 = reshape(sta1,nkt,[]); sps2 = sum(reshape(sps(:,2),[],slen),1)'; % rebinned spike train sta2 = simpleSTC(Stim,sps2,nkt); % Compute STA 2 sta2 = reshape(sta2,nkt,[]); % Initialize param struct for fitting gg0 = makeFittingStruct_GLM(dtStim,dtSp); % Initialize params for fitting struct % Initialize fields (using h and k bases computed above) gg0.ktbas = ktbas; % k basis gg0.ihbas = ihbas; % h self-coupling basis gg0.ihbas2 = ihbas; % h coupling-filter basis nktbasis = size(ktbas,2); % number of basis vectors in k basis nhbasis = size(ihbas,2); % number of basis vectors in h basis gg0.kt = 0.1*(ktbas\sta1); % initial params from scaled-down sta gg0.k = gg0.ktbas*gg0.kt; % initial setting of k filter gg0.ihw = zeros(nhbasis,1); % params for self-coupling filter gg0.ihw2 = zeros(nhbasis,1); % params for cross-coupling filter gg0.ih = [gg0.ihbas*gg0.ihw gg0.ihbas2*gg0.ihw2]; gg0.iht = iht; gg0.dc = 0; % Initialize dc term to zero gg0.couplednums = 2; % cell numbers of cells coupled to this one % Set spike responses for cell 1 and coupled cell gg0.sps = sps(:,1); gg0.sps2 = sps(:,2); % Compute initial value of negative log-likelihood (just to inspect) [neglogli0,rr] = neglogli_GLM(gg0,Stim); % Do ML fitting fprintf('Fitting neuron 1: initial neglogli0 = %.3f\n', neglogli0); opts = {'display', 'iter', 'maxiter', 100}; [gg1, neglogli1] = MLfit_GLM(gg0,Stim,opts); % do ML (requires optimization toolbox) %% ===== 5. Fit cell #2 (with coupling from cell #1) ================== gg0b = gg0; % initial parameters for fitting gg0b.sps = sps(:,2); % cell 2 spikes gg0b.sps2 = sps(:,1); % spike trains from coupled cells gg0.kt = 0.1*(ktbas\sta2); % initial params from scaled-down sta gg0b.k = gg0b.ktbas*gg0b.kt; % Project STA onto basis for fitting gg0.couplednums = 1; % number of cell coupled to this one % Compute initial value of negative log-likelihood (just to inspect) [neglogli0b] = neglogli_GLM(gg0b,Stim); % initial value of negative logli % Do ML fitting fprintf('Fitting neuron 2: initial neglogli = %.3f\n', neglogli0b); [gg2, neglogli2] = MLfit_GLM(gg0b,Stim,opts); % do ML (requires optimization toolbox) %% ===== 6. Plot fits ============================================= %% ttk = (-nkt+1:0)*dtStim; % time indices for k subplot(221); % Stim filters cell 1 % --------------- plot(ttk, ggsim1.k(:,1), 'k', ttk, gg1.k, 'r'); title('Cell 1: k filter'); legend('true', 'estim', 'location', 'northwest'); subplot(223); % Stim filters cell 2 % --------------- plot(ttk, ggsim2.k, 'k', ttk, gg2.k, 'r'); title('Cell 2: k filter'); xlabel('time (s)') subplot(222); % --Spike filters cell 1 % ------------- plot(ggsim.iht, (ggsim.ih(:,:,1)), gg1.iht, (gg1.ih), '--'); title('exponentiated incoming h filters'); legend('true h11', 'true h21', 'estim h11', 'estim h21'); axis tight; subplot(224); % --Spike filters cell 2 % ------------- plot(ggsim.iht, (ggsim.ih(:,:,2)), gg2.iht, (gg2.ih), '--'); title('exponentiated incoming h filters'); axis tight; xlabel('time (s)') legend('true h22', 'true h12', 'estim h22', 'estim h12');
MATLAB
2D
pillowlab/GLMspiketools
demos/demo2b_GLM_spatialStim_Regularized.m
.m
8,074
224
% demo2b_GLM_spatialStim_Regularized.m % % Test code for simulating and fitting the GLM with a 2D stimulus filter % (time x 1D space), regularized with a Gaussian prior. % Make sure paths are set (assumes this script called from 'demos' directory) cd ..; setpaths_GLMspiketools; cd demos/ %% 1. Set parameters and display for GLM ============= % dtStim = .01; % Bin size for stimulus (in seconds). (Equiv to 100Hz frame rate) dtSp = .001; % Bin size for simulating model & computing likelihood (must evenly divide dtStim); % Make a temporal filter nkt = 20; % Number of time bins in stimulus filter k kt = (normpdf(1:nkt,3*nkt/4,1.5)-.5*normpdf(1:nkt,nkt/2,3))'; % Make a spatial filter; nkx = 10; xxk = (1:1:nkx)'; % pixel locations ttk = dtStim*(-nkt+1:0)'; % time bins for filter kx = 1./sqrt(2*pi*4).*exp(-(xxk-nkx/2).^2/5); ktrue = kt*kx'; % Make space-time separable filter ktrue = ktrue./norm(ktrue(:)); % Insert into glm structure (created with default history filter) ggsim = makeSimStruct_GLM(nkt,dtStim,dtSp); % Create GLM structure with default params ggsim.k = ktrue; % Insert into simulation struct ggsim.dc = 3; % === Make Fig: model params ======================= clf; subplot(3,3,[1 4]); % ------------------------------------------ plot(kt,ttk); axis tight; set(gca, 'ydir', 'reverse'); ylabel('time (frames)'); subplot(3,3,[2 3 5 6]); % -------------------------------------- imagesc(xxk,ttk,ggsim.k); colormap gray; axis off; title('stimulus kernel k'); subplot(3,3,8:9); % ---------------------------------------- plot(xxk,kx); axis tight; set(gca, 'xlim', [.5 nkx+.5]); xlabel('space (pixels)'); %% 2. Generate some training data ======================================== % generate stimulus slen = 10000; % Stimulus length (frames); More samples gives better fit gfilt = normpdf(-3:3, 0, .8); Stim = conv2(randn(slen,nkx),gfilt,'same'); % convolve in space Stim = conv2(Stim,gfilt','same'); % convolve in time [tsp,sps,Itot,Istm] = simGLM(ggsim,Stim); % run model nsp = length(tsp); rlen = length(sps); % --- Make plot of first 0.5 seconds of data -------- tlen = 0.5; ttstim = dtStim:dtStim:tlen; iistim = 1:length(ttstim); ttspk = dtSp:dtSp:tlen; iispk = 1:length(ttspk); spinds = sps(iispk)>0; subplot(311); imagesc([0 tlen], [1 nkx], Stim(iistim,:)'); title('stimulus'); ylabel('pixel'); subplot(312); semilogy(ttspk,exp(Itot(iispk)),ttspk(spinds), exp(Itot(spinds)), 'ko'); ylabel('spike rate (sp/s)'); title('conditional intensity (and spikes)'); subplot(313); Isp = Itot-Istm; % total spike-history filter output plot(ttspk,Istm(iispk), ttspk,Isp(iispk)); axis tight; legend('k output', 'h output'); xlabel('time (s)'); ylabel('log intensity'); title('filter outputs'); %% 3. Fit GLM with max likelihood % Compute the STA sps_coarse = sum(reshape(sps,[],slen),1)'; % bin spikes in bins the size of stimulus sta0 = simpleSTC(Stim,sps_coarse,nkt); % Compute STA sta = reshape(sta0,nkt,[]); % reshape it to match dimensions of true filter exptmask= []; % Not currently supported! % Initialize params for fitting -------------- % Set params for fitting, including bases nkbasis = 8; % number of basis vectors for representing k nhbasis = 8; % number of basis vectors for representing h hpeakFinal = .1; % time of peak of last basis vector for h gg0 = makeFittingStruct_GLM(dtStim,dtSp,nkt,nkbasis,sta,nhbasis,hpeakFinal); gg0.sps = sps; % Insert binned spike train into fitting struct gg0.mask = exptmask; % insert mask (optional) gg0.ihw = randn(size(gg0.ihw))*1; % initialize spike-history weights randomly % Compute conditional intensity at initial parameters [negloglival0,rr] = neglogli_GLM(gg0,Stim); fprintf('Initial negative log-likelihood: %.5f\n', negloglival0); % Do ML estimation of model params opts = {'display', 'iter', 'maxiter', 1000}; [gg1, negloglival1] = MLfit_GLM(gg0,Stim,opts); % do ML (requires optimization toolbox) %% 4. Fit GLM with iid Gaussian ("ridge regression") prior. % Divide data into training and test trainfrac = 0.8; ntrain = ceil(trainfrac*slen); ntrainsps = ceil(trainfrac*rlen); stimtrain = Stim(1:ntrain,:); stimtest = Stim(ntrain+1:end,:); spstrain = sps(1:ntrainsps); spstest = sps(ntrainsps+1:end); % Set prior covariance matrix --------------- nparams = numel(gg0.kt); Ceye = speye(nparams); % Make grid of lambda (ridge parameter) values ----- lamvals = 2.^(-1:10); nlam = length(lamvals); testlogli = zeros(nlam,1); trainlogli = zeros(nlam,1); ggridge = cell(nlam,1); % Initialize params -------- gg0_train = gg0; gg0_train.sps = spstrain; % Find MAP estimate for each value of ridge parameter fprintf('====Doing grid search for ridge parameter====\n'); for jj = 1:nlam fprintf('lambda = %.1f\n',lamvals(jj)); Cinv = lamvals(jj)*Ceye; % inverse covariance of prior % Do MAP estimation of model params [ggridge{jj}, negloglival] = MAPfit_GLM(gg0_train,stimtrain,Cinv,opts); % do ML (requires optimization toolbox) trainlogli(jj) = -negloglival; % Compute test log-likelihood gg2tst = ggridge{jj}; gg2tst.sps = spstest; testlogli(jj) = -neglogli_GLM(gg2tst,stimtest); end % Plot test log likelihood clf; semilogx(lamvals,testlogli, '-o'); xlabel('lambda'); ylabel('test log-likelihood'); title('ridge prior'); % Select best ridge param and fit on all data. [~,imax] = max(testlogli); [gg2,neglogli2] = MAPfit_GLM(gg1,Stim,lamvals(imax)*Cinv,opts); %% 5. Fit GLM with smoothing prior. % Make matrices to compute column and row pairwise differences Dx = spdiags(ones(nkx,1)*[-1 1],0:1,nkx-1,nkx); Dt = spdiags(ones(nkt,1)*[-1 1],0:1,nkt-1,nkt); Dtbas = Dt*gg1.ktbas; Mt = kron(speye(nkx),Dtbas'*Dtbas); % computes squared diffs along columns Mx = kron(Dx,gg1.ktbas); % computes squared diffs along row Mx = Mx'*Mx; C0 = Mt + Mx; % default covariance matrix % Make grid of lambda (smoothing parameter) values ----- lamvals_sm = 2.^(0:13); nlam = length(lamvals_sm); testlogli_sm = zeros(nlam,1); trainlogli_sm = zeros(nlam,1); ggsmooth = cell(nlam,1); % Find MAP estimate for each value of ridge parameter for jj = 1:nlam Cinv = lamvals_sm(jj)*C0; % inverse covariance of prior % Do MAP estimation of model params [ggsmooth{jj}, negloglival] = MAPfit_GLM(gg0_train,stimtrain,Cinv,opts); % do ML (requires optimization toolbox) trainlogli_sm(jj) = -negloglival; % Compute test log-likelihood gg3tst = ggsmooth{jj}; gg3tst.sps = spstest; testlogli_sm(jj) = -neglogli_GLM(gg3tst,stimtest); end % Plot test log likelihood clf; semilogx(lamvals_sm,testlogli_sm, '-o'); xlabel('lambda'); ylabel('test log-likelihood'); title('smoothing prior'); % Select best ridge param and fit on all data. [~,imax] = max(testlogli_sm); [gg3,neglogli3] = MAPfit_GLM(gg2,Stim,lamvals_sm(imax)*C0,opts); %% 6. Plot results ==================== subplot(231); % True filter % --------------- imagesc(ggsim.k); colormap gray; title('True Filter');ylabel('time'); subplot(232); % sta % ------------------------ imagesc(gg1.k); title('ML estimate'); ylabel('time'); subplot(234); % sta-projection % --------------- imagesc(gg2.k); title('MAP: ridge prior'); xlabel('space'); ylabel('time'); subplot(235); % estimated filter % --------------- imagesc(gg3.k); title('MAP: smoothing prior'); xlabel('space'); subplot(236); % ---------------------------------- plot(ggsim.iht,exp(ggsim.ih),'k--', gg1.iht,exp(gg1.ihbas*gg1.ihw),... gg2.iht, exp(gg2.ihbas*gg2.ihw), ... gg3.iht, exp(gg3.ihbas*gg3.ihw) ... ); axis tight; title('post-spike kernel'); xlabel('time after spike (s)'); legend('true','ML','ridge','smooth'); r2fun = @(k)(1-sum((k(:)-ktrue(:)).^2)./sum(ktrue(:).^2)); % error function fprintf('\nK filter R^2:\n ----------\n ML = %9.3f\n Ridge = %6.3f\n Smooth = %4.3f\n',... r2fun(gg1.k), r2fun(gg2.k),r2fun(gg3.k));
MATLAB
2D
pillowlab/GLMspiketools
demos/demo4_GLM_3CoupledNeurons.m
.m
9,026
220
% testscript3_GLM_coupled.m % % Demo script for simulating and fitting a coupled GLM (2 neurons). % % Notes: % - Fitting code uses same functions as for single-cell responses. % - Simulation code requires new structures / functions % (due to the need to pass activity between neurons) % Make sure paths are set (assumes this script called from 'demos' directory) cd ..; setpaths_GLMspiketools; cd demos/ %% ===== 1. Set parameters for simulating a GLM ============ % dtStim = .001; % Bin size for simulating model & computing likelihood (in units of stimulus frames) dtSp = .001; % Bin size for simulating model & computing likelihood (must evenly divide dtStim); nkt = 1; % Number of time bins in stimulus filter k %% FIX THIS SO IT DOESN'T MAKE NAN gg = makeSimStruct_GLM(nkt,dtStim,dtSp); % Create GLM struct with default params %% Set up basis for self-coupling filters % Make basis for self-coupling term ihbasprs.ncols = 3; % number of basis vectors ihbasprs.hpeaks = [.002, .005]; % peak of 1st and last basis vector ihbasprs.b = .001; % scaling (smaller -> more logarithmic scaling) ihbasprs.absref = .002; % absolute refractory period basis vector (optional) % Make basis [iht,ihbas,ihbasis] = makeBasis_PostSpike(ihbasprs,dtSp); nht = length(iht); % number of bins % Make basis for cross-coupling term ihbasprs2.ncols = 1; % number of basis vectors ihbasprs2.hpeaks = [0.001,.005]; % put peak at 10ms and "effective" 1st peak at 0 ihbasprs2.b = .002; % smaller -> more logarithmic scaling ihbasprs2.absref = []; % no abs-refracotry period for this one % Make basis [iht2,ihbas2,ihbasis2] = makeBasis_PostSpike(ihbasprs2,dtSp); nht2 = length(iht2); % pad to put them into the same time bins, if necessary if nht2>nht % padd ih1 with zeros iht = iht2; zz = zeros(nht2-nht,ihbasprs.ncols); ihbas = [ihbas;zz]; ihbasis = [ihbasis;zz]; nht=nht2; elseif nht2<nht % padd ih1 with zeros iht2 = iht; zz = zeros(nht-nht2,ihbasprs2.ncols); ihbas2 = [ihbas2;zz]; ihbasis2 = [ihbasis2;zz]; nht2=nht; end % plot it subplot(211); plot(iht, ihbasis, ihbasprs.hpeaks, 1, '*', 'linewidth', 2); xlabel('time after spike (ms)'); title('post-spike basis'); subplot(212); plot(iht2, ihbasis2, ihbasprs2.hpeaks, 1, '*', 'linewidth', 2); xlabel('time after spike (ms)'); title('coupling basis'); %% Set self-coupling weights wself = [-5; .2; -.15]; % weights for self-coupling term ihself = ihbasis*wself; % self-coupling filter wcpl = 0.5; % weights for cross-coupling term ihcpl = ihbasis2*wcpl; % cross-coupling filter clf; plot(iht, exp(ihself), iht, exp(ihcpl), iht, iht*0+1, 'k--'); legend('self-coupling', 'cross-coupling'); xlabel('time lag (s)'); ylabel('gain (sp/s)'); %% Set up multi-neuron GLM nneur = 3; k = [.85 .9 0.7]; % stimulus weights gg.k = permute(k,[1,3,2]); % stimulus weights gg.iht = iht; gg.ih = zeros(nht,nneur,nneur); gg.ih(:,:,1) = [ihself, 1.25*ihcpl, -1*ihcpl]; % input weights to neuron 1 gg.ih(:,:,2) = [1.8*ihcpl, 1.5*ihself, -.5*ihcpl]; % input weights to neuron 2 gg.ih(:,:,3) = [1.5*ihcpl, .5*ihcpl, ihself]; % input weights to neuron 3 %% ===== 2. Run short simulation for visualization purposes ========= % slen = 50; % Stimulus length (frames) & width (# pixels) swid = 1; % width of stimulus stimsd = 1; % contrast of stimulus Stim = stimsd*randn(slen,swid); % Gaussian white noise stimulus [tsp,~,Itot,Istm] = simGLM(gg, Stim); % Simulate GLM response Isp = Itot-Istm; % net spike-history output % ==== Plot some traces of simulated response ======== tt = (dtSp:dtSp:slen*dtStim)'; subplot(131) % % ==== neuron 1 =========== plot(tt, Istm(:,1), 'k', tt, Isp(:,1), 'r', tt, tt*0, 'k--', ... tsp{1}, ones(size(tsp{1})), 'ro'); title('cell 1'); axis tight; xlabel('time (s)'); ylabel('filter output'); legend('stim filter + dc', 'spike-hist filter'); subplot(132) % ==== neuron 2 =========== plot(tt, Istm(:,2), 'k', tt, Isp(:,2), 'r', tt, tt*0, 'k--', ... tsp{2}, ones(size(tsp{2})), 'ro'); title('cell 2'); axis tight; xlabel('time (s)'); subplot(133) % ==== neuron 3 =========== plot(tt, Istm(:,3), 'k', tt, Isp(:,3), 'r', tt, tt*0, 'k--', ... tsp{3}, ones(size(tsp{3})), 'ro'); title('cell 3'); axis tight; xlabel('time (s)'); %% ===== 3. Generate some training data =============================== %% slen = 2e5; % Stimulus length (in bins); Stim = stimsd*randn(slen,swid); % Gaussian white noise stimulus [tsp,sps,Itot,ispk] = simGLM(gg,Stim); % run model %% ===== 4. Fit cell #1 (with coupling from cell #2 and #3) =================== %% % Initialize param struct for fitting gg1in = makeFittingStruct_GLM(dtStim,dtSp); % Initialize params for fitting struct % Initialize fields (using h bases computed above) gg1in.ktbas = 1; % k basis gg1in.ihbas = ihbas; % h self-coupling basis gg1in.ihbas2 = ihbas2; % h coupling-filter basis nktbasis = 1; % number of basis vectors in k basis nhbasis = size(ihbas,2); % number of basis vectors in h basis nhbasis2 = size(ihbas2,2); % number of basis vectors in h basis gg1in.kt = 1; % initial params from scaled-down sta gg1in.k = 1; % initial setting of k filter gg1in.ihw = zeros(nhbasis,1); % init params for self-coupling filter gg1in.ihw2 = zeros(nhbasis2,nneur-1); % init params for cross-coupling filter gg1in.ih = [gg1in.ihbas*gg1in.ihw gg1in.ihbas2*gg1in.ihw2]; gg1in.iht = iht; gg1in.dc = 0; % Initialize dc term to zero % Set fields for fitting cell #1 couplednums = [2 3]; % the cells coupled to this one gg1in.couplednums = couplednums; % cell numbers of cells coupled to this one gg1in.sps = sps(:,1); % Set spike responses for cell 1 gg1in.sps2 = sps(:,couplednums); % spikes from coupled cells % Compute initial value of negative log-likelihood (just to inspect) [neglogli0,rr] = neglogli_GLM(gg1in,Stim); % Do ML fitting fprintf('Fitting neuron 1: initial neglogli0 = %.3f\n', neglogli0); opts = {'display', 'iter', 'maxiter', 100}; [gg1, neglogli1] = MLfit_GLM(gg1in,Stim,opts); % do ML (requires optimization toolbox) %% ===== 5. Fit cell #2 (with coupling from cell #1 and #3) ================== cellnum = 2; couplednums = setdiff(1:3, cellnum); % the cells coupled to this one gg2in = gg1in; % initial parameters for fitting gg2in.sps = sps(:,cellnum); % cell 2 spikes gg2in.sps2 = sps(:,couplednums); % spike trains from coupled cells gg2in.couplednums = couplednums; % cells coupled to this one % Do ML fitting fprintf('Fitting neuron 2\n'); [gg2, neglogli2] = MLfit_GLM(gg2in,Stim,opts); % do ML (requires optimization toolbox) %% ===== 6. Fit cell #3 (with coupling from cell #1 and #2) ================== cellnum = 3; couplednums = setdiff(1:3, cellnum); gg3in = gg1in; % initial parameters for fitting gg3in.sps = sps(:,cellnum); % cell 3 spikes gg3in.sps2 = sps(:,couplednums); % spike trains from coupled cells gg3in.couplednums = couplednums; % cells coupled to this one % Do ML fitting fprintf('Fitting neuron 3\n'); [gg3, neglogli3] = MLfit_GLM(gg3in,Stim,opts); % do ML (requires optimization toolbox) %% ===== 6. Plot fits ============================================= %% colors = get(gca,'colororder'); set(gcf,'defaultAxesColorOrder',colors(1:3,:)); % use only 3 colors lw = 2; % linewidth ymax = max(exp([gg1.ih(:);gg2.ih(:);gg3.ih(:);gg.ih(:)])); % max of y range subplot(131); % --Spike filters cell 1 % ------------- plot(gg.iht, exp(gg.ih(:,:,1)), gg1.iht, exp((gg1.ih)), '--', 'linewidth', lw); hold on; plot(gg.iht, gg.iht*0+1, 'k'); hold off; legend('true h11', 'true h21', 'true h31', 'estim h11', 'estim h21', 'estim h31', 'location', 'southeast'); title('incoming filters: cell 1'); axis tight; set(gca,'ylim',[0,ymax]); ylabel('gain (sp/s)'); xlabel('time after spike (s)'); subplot(132); % --Spike filters cell 2 % ------------- plot(gg.iht, exp(gg.ih(:,[2 1 3],2)), gg2.iht, exp(gg2.ih), '--', 'linewidth', lw); hold on; plot(gg.iht, gg.iht*0+1, 'k'); hold off; legend('true h22', 'true h12', 'true h32', 'estim h22', 'estim h12', 'estim h32', 'location', 'southeast'); title('cell 2'); axis tight; set(gca,'ylim',[0,ymax]); subplot(133); % --Spike filters cell 3 % ------------- plot(gg.iht, exp(gg.ih(:,[3 1 2],3)), gg3.iht, exp(gg3.ih), '--', 'linewidth', lw); hold on; plot(gg.iht, gg.iht*0+1, 'k'); hold off; legend('true h33', 'true h13', 'true h23', 'estim h22', 'estim h13', 'estim h23', 'location', 'southeast'); title('cell 3'); axis tight; set(gca,'ylim',[0,ymax]); % Print out true and recovered params: fprintf('\n------------------\n'); fprintf('True stim weights: %.2f, %.2f, %.2f\n', squeeze(gg.k)); fprintf(' Est stim weights: %.2f, %.2f, %.2f\n\n', gg1.k,gg2.k,gg3.k); % Print out true and recovered params: fprintf('True dc: %.2f, %.2f, %.2f\n', gg.dc*[1 1 1]); fprintf(' Est dc: %.2f, %.2f, %.2f\n', gg1.dc,gg2.dc,gg3.dc);
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/MLfit_GLMbi.m
.m
1,644
49
function [gg,fval,H,Xstruct] = MLfit_GLMbi(gg,Stim,optimArgs) % [gg,fval,H,Xstruct] = MLfit_GLMbi(gg,Stim,optimArgs); % % Computes the ML estimate for GLM params, using grad and hessians. % Assumes bilinear parametrization of space-time filter. % % Inputs: % gg = param struct % Stim = stimulus % optimArgs = cell array of optimization params (optional) % % Outputs: % ggnew = new param struct (with ML params); % fval = negative log-likelihood at ML estimate % H = Hessian of negative log-likelihood at ML estimate % Xstruct = structure with design matrices for spike-hist and stim terms % Set optimization parameters algopts = getFminOptsForVersion(version); if nargin > 2, opts = optimset(algopts{:}, optimArgs{:}); else, opts = optimset(algopts{:}); end % --- Create design matrix extract initial params from gg ---------------- [prs0,Xstruct] = setupfitting_GLM(gg,Stim); % minimize negative log likelihood lfunc = @(prs)Loss_GLMbi_logli(prs,Xstruct); % loss function for exponential nonlinearity [prs,fval] = fminunc(lfunc,prs0,opts); % optimize negative log-likelihood for prs % Compute Hessian at maximum, if requested if nargout > 2 [fval,~,H] = Loss_GLMbi_logli(prs); end % Put returned vals back into param structure ------ gg = reinsertFitPrs_GLMbi(gg,prs,Xstruct); % %---------------------------------------------------- % Optional debugging code % %---------------------------------------------------- % % % ------ Check analytic gradients, Hessians ------- % DerivCheck(lfunc,prs0,opts); % HessCheck_Elts(lfunc, [1 12],prs0,opts); % tic; [lival,J,H]=lfunc(prs0,Xstruct); toc;
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/MAPfit_GLMbi_coordascent.m
.m
4,334
133
function [gg,neglogli,H,neglogp] = MAPfit_GLMbi_coordascent(gg,Stim,Cxinv,Ctinv,maxiter,ftol,optimArgs) % [gg,neglogli,H,neglogp] = MAPfit_GLMbi_coordascent(gg,Stim,Cxinv,Ctinv,maxiter,ftol,optimArgs) % % Computes the MAP estimate for GLM params, with gradient and hessians, via % coordinate ascent, for bilinear (low rank) parametrization of space-time filter. % % Inputs: % gg = param struct % Stim = stimulus % Cxinv = inverse covariance for spatial params % Ctinv = inverse covariance for temporal params % maxiter = maximum number of coordinate ascent steps % ftol = tolerance for stopping based on function improvement % optimArgs = cell array of optimization params (optional) % % Outputs: % ggnew = new param struct (with ML params); % fval = negative log-likelihood at ML estimate % H = Hessian of negative log-likelihood at ML estimate % Set maxiter if necessary if (nargin <= 3) || isempty(maxiter) maxiter = 50; % maximum number of coordinate ascent steps end if (nargin <=4) || isempty(ftol) ftol = .001; % tolerance for stopping end % Set optimization parameters if nargin < 6 optimArgs = []; end % create struct for spatial optimization ggx = gg; ggx.ktype = 'linear'; % set type to linear (from bilinear) ggx.ktbas = 1; ggx.ktbasprs = []; % remove temporal basis ggx.kt = gg.kx(:)'; ggx.k = ggx.kt; % create struct for temporal optimization ggt = gg; ggt.ktype = 'linear'; % set type to linear (from bilinear) ggt.kt = gg.kt; ggt.k = ggt.ktbas*ggt.kt; % Initialize spatial stimulus [nt,nkx] = size(Stim); krank = gg.krank; xStim = zeros(nt,krank*nkx); % initialize spatial stimulus % compute initial log-likelihood neglogli0 = neglogli_GLM(gg,Stim); % Compute logli of initial params dlogp = inf; % initialize change in log-posterior jjiter = 0; % initialize counter % compute initial penalties (negative log prior) kt = gg.kt; % temporal params kx = gg.kx; % spatial params krank = size(kt,2); % rank nlpx = .5*kx'*Cxinv*kx; nlpt = .5*kt'*Ctinv*kt; nlp = trace(nlpx*nlpt); neglogp0 = neglogli0+nlp; % initial log-posterior fprintf('Initial smoothing penalty: %.2f\n', nlp); % Do coordinate ascent until STOP while (jjiter<maxiter) && dlogp>ftol % ---- Update temporal params ----- fprintf('Iter #%d: Updating t params\n', jjiter); tStim = Stim*reshape(ggx.k',[],gg.krank); ggt.dc = ggx.dc; % update dc param [ggt,tneglogli] = MAPfit_GLM(ggt,tStim,kron(nlpx,Ctinv),optimArgs); nlpt = .5*ggt.kt'*Ctinv*ggt.kt; nlp = trace(nlpx*nlpt); fprintf(' dlogp = %.4f (penalty=%.2f)\n', neglogp0-(tneglogli+nlp), nlp); % Convolve stimulus with temporal filters for irank = 1:krank for icol = 1:nkx xStim(:,icol+(irank-1)*nkx) = sameconv(Stim(:,icol),ggt.k(:,irank)); end end % ---- Update spatial params ---- fprintf('Iter #%d: Updating x params\n', jjiter); ggx.dc = ggt.dc; % update dc param [ggx,xneglogli] = MAPfit_GLM(ggx,xStim,kron(nlpt,Cxinv),optimArgs); nlpx = .5*reshape(ggx.k,[],krank)'*Cxinv*reshape(ggx.k(:),[],krank); nlp = trace(nlpx*nlpt); neglogp = xneglogli+nlp; fprintf(' dlogp = %.4f (penalty=%.2f)\n', neglogp0-neglogp,nlp); % Update iters jjiter = jjiter+1; % counter dlogp = neglogp0-neglogp; % change in log-likelihood neglogp0 = neglogp; end fprintf('Finished coordinate ascent: %d iterations (dlogp=%.6f)\n',jjiter,dlogp); % Compute conditional Hessians, if desired if nargout > 2 % Compute Hessian for time components tStim = Stim*(ggx.k'); ggt.dc = ggx.dc; % update dc param [ggt,~,Ht] = MAPfit_GLM(ggt,tStim,Ctinv,optimArgs); nlpt = .5*ggt.kt(:)'*Ctinv*ggt.kt(:); % Compute Hessian for space components for irank = 1:krank for icol = 1:nkx xStim(:,icol+(irank-1)*nkx) = sameconv(Stim(:,icol),ggt.k(:,irank)); end end ggx.dc = ggt.dc; % update dc param [ggx,xneglogli,Hx] = MAPfit_GLM(ggx,xStim,Cxinv,optimArgs); nlpx = .5*ggx.k(:)'*Cxinv*ggx.k(:); neglogp = xneglogli+nlpt+nlpx; H = {Ht,Hx}; % Hessians for t and x components end % Update params of bilinear model gg.dc = ggx.dc; gg.kt = ggt.kt; gg.kx = reshape(ggx.k',[],gg.krank); gg.k = (gg.ktbas*gg.kt)*gg.kx'; neglogli = xneglogli;
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/reinsertFitPrs_GLMbi.m
.m
759
30
function gg = reinsertFitPrs_GLMbi(gg,prs,Xstruct) % gg = reinsertFitPrs_GLMbi(gg,prs); % % After fitting, reinsert params into param structure % Put returned vals back into param structure ------ krank = Xstruct.krank; nktprs = Xstruct.nkt*krank; nkxprs = Xstruct.nkx*krank; nktot = nktprs+nkxprs; nh = Xstruct.nh; nh2 = Xstruct.nh2; % Insert params into struct gg.kt = reshape(prs(1:nktprs),[],krank); gg.kx = reshape(prs(nktprs+1:nktprs+nkxprs),[],krank); gg.dc = prs(nktot+1); gg.ihw = reshape(prs(nktot+2:nktot+nh+1), nh,1); gg.ihw2 = reshape(prs(nktot+nh+2:end), nh2, []); gg.k = (gg.ktbas*gg.kt)*(gg.kxbas*gg.kx)'; if isempty(gg.ihw) gg.ihbas = []; end if isempty(gg.ihw2) gg.ihbas2 = []; end gg.ih = [gg.ihbas*gg.ihw, gg.ihbas2*gg.ihw2];
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/reinsertFitPrs_GLM.m
.m
705
28
function gg = reinsertFitPrs_GLM(gg,prs,Xstruct) % gg = reinsertFitPrs_GLM(gg,prs,Xstruct) % % After fitting, reinsert params into GLM param structure % Extract relevant size information nkt = Xstruct.nkt; nkx = Xstruct.nkx; nktot = nkt*nkx; nh = Xstruct.nh; nh2 = Xstruct.nh2; % Insert params gg.kt = reshape(prs(1:nktot),nkt,nkx); gg.k = gg.ktbas*gg.kt; gg.dc = prs(nktot+1); gg.ihw = reshape(prs(nktot+2:nktot+nh+1), nh,1); gg.ihw2 = reshape(prs(nktot+nh+2:end), nh2, []); % Ensure these bases are 'empty' if no spike history filters if isempty(gg.ihw), gg.ihbas = []; end if isempty(gg.ihw2), gg.ihbas2 = []; end % Insert spike-history filters gg.ih = [gg.ihbas*gg.ihw, gg.ihbas2*gg.ihw2];
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/initfit_stimDesignMat_bi.m
.m
1,968
48
function Xstruct = initfit_stimDesignMat_bi(gg,Stim) % Xstruct = initfit_stimDesignMat_bi(gg,Stim) % % Initialize parameters relating to stimulus design matrix for bilinearly % parametrized GLM % ---- Set up filter and stim processing params ------------------- nkx = size(gg.kxbas,2); % # params per spatial vector nkt = size(gg.ktbas,2); % # time params per stim pixel (# t params) ncols = nkx*nkt; % number of columns in design matrix [slen,swid] = size(Stim); % size of stimulus upsampfactor = (gg.dtStim/gg.dtSp); % number of times by which spikes more finely sampled than stim rlen = slen*upsampfactor; % length of spike train vector % ---- Check size of filter and width of stimulus ---------- assert(nkx == swid,'Mismatch between stim width and kernel width'); % ---- Convolve stimulus with spatial and temporal bases ----- xfltStim = Stim*gg.kxbas; % stimulus filtered with spatial basis % ---- Allocated space for design matrix ------ try Xstruct.Xstim = zeros(slen,ncols); % catch fprintf('\nERROR: Stimulus too for joint optimization!\n'); fprintf('Use ''MLfit_GLMbi_coordascent.m'' instead\n\n'); error('Out of Memory Error in initfit_stimDesignMat_bi.m'); end % ---- Build stimulus design matrix ------------- for i = 1:nkx for j = 1:nkt Xstruct.Xstim(:,(i-1)*nkt+j) = sameconv(xfltStim(:,i),gg.ktbas(:,j)); end end % ---- Set up stim processing params ---------- Xstruct.nkx = nkx; % # stimulus spatial stimulus pixels Xstruct.nkt = nkt; % # time bins in stim filter Xstruct.krank = gg.krank; % stim filter rank Xstruct.slen = slen; % Total stimulus length (coarse bins) Xstruct.rlen = rlen; % Total spike-train bins (fine bins) Xstruct.upsampfactor = upsampfactor; % rlen / slen Xstruct.Minterp = kron(speye(slen),ones(upsampfactor,1)); % rlen x slen matrix for upsampling and downsampling Xstruct.dtStim = gg.dtStim; % time bin size for stimulus Xstruct.dtSp = gg.dtSp; % time bins size for spike train
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/neglogli_GLM.m
.m
3,083
93
function [neglogli,rr,tt,Itot,Istm,Ih,Icpl] = neglogli_GLM(gg,Stim) % [neglogli,rr,tt,Itot,Istm,Ih,Icpl] = neglogli_GLM(gg,Stim) % % Compute glm model negative log-likelihood given the parameters in gg, % % Inputs: gg = param object % fields: k - stimulus kernel % ih - post-spike current % dc - dc current injection % nlfun - nonlinearity % dt - time bin size % Stim = stimulus % % Outputs: % neglogli = negative log-likelihood of spike trains % rr = conditional intensity (in expected spikes /sec) % tt = time bins for for conditional intensity % Itot = sum of all filter outputs % Istm = sum of stim filter output and dc term % Ih = output of neuron's own spike-history filter % Icpl = matrix of coupling current inputs % ---- Extract params from glm struct --------------- k = gg.k; dc = gg.dc; ih = gg.ih; dt = gg.dtSp; % time bin size for spikes upsampfactor = gg.dtStim/dt; % number of spike bins per Stim bin slen = size(Stim,1); rlen = size(gg.sps,1); % Check that upSampFactor is an integer assert(mod(upsampfactor,1) == 0, 'dtStim / dtSp must be an integer'); % Check factor relating size of stim and binned spike train assert(slen*upsampfactor==rlen,'Spike train length must be an even multiple of stim length'); % ---- Compute filtered resp to stimulus ----------------------------- I0 = sameconv(Stim,k); Itot = kron(I0,ones(upsampfactor,1)) + dc; % -------------- Compute net h current -------------------------------- % Check if post-spike filters are present if isempty(gg.ihw), gg.ihbas = []; end if isempty(gg.ihw2), gg.ihbas2 = []; end % Compute convolution of post-spike filters with spike history nCoupled = length(gg.couplednums); % # cells coupled to this one if ~isempty(ih) Itot = Itot + spikefilt(gg.sps,ih(:,1)); % self-coupling filter for j = 1:nCoupled % coupling filters from other neurons Itot = Itot + spikefilt(gg.sps2(:,j),ih(:,j+1)); end end rr = gg.nlfun(Itot); % Conditional intensity % -- Compute negative log-likelihood from conditional intensity ------ bmask = initfit_mask(gg.mask,dt,rlen); % bins to use for likelihood calc rrmask = rr(bmask); % rate only on timee bins in mask trm1 = sum(rrmask)*dt; % non-spiking term trm2 = -sum(log(rrmask(gg.sps(bmask)>0))); % Spiking term neglogli = trm1 + trm2; % ====== OPTIONAL OUTPUT ARGS =================================== % time indices if nargout > 2, tt = (1:rlen)'*dt; end % Stim filter output if nargout > 4, Istm = kron(I0,ones(upsampfactor,1)) + dc; end % Spike-history filter output if nargout > 5 Ih = zeros(length(Itot),1); if ~isempty(ih), Ih = spikefilt(gg.sps,ih(:,1)); end end % Coupling filter outputs if nargout > 6 Icpl = zeros(length(Itot),size(ih,2)-1); if ~isempty(ih) for j = 1:nCoupled Icpl(:,j) = spikefilt(gg.sps2(:,j),ih(:,j+1)); end end end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/Loss_GLMbi_logli.m
.m
6,263
190
function [logli, dL, H] = Loss_GLMbi_logli(prs,Xstruct) % [logli, dL, H] = Loss_GLMbi_logli(prs,Xstruct) % % Compute negative log-likelXXspood of data undr the GLM model, with bilinear % parametrization of the input current % % Uses arbitrary nonlinearity 'nlfun' instead of exponential % % Inputs: % prs = [ktprs - weights for stimulus time kernel % kxprs - weights for stim space kernel % dc - dc current injection % XXspprs - weights on post-spike current]; % % Outputs: % logli = negative log likelXXspood of spike train % dL = gradient with respect to prs % H = hessian % Extract some size information from Xstruct nkt = Xstruct.nkt; % # of basis vectors for spatial k basis nkx = Xstruct.nkx; % # of basis vectors for spatial k basis krank = Xstruct.krank; % rank of stim filter nktprs = nkt*krank; % total # params for temporal filters nkxprs = nkx*krank; % total # params for spatial filters % Unpack GLM prs; ktprs = prs(1:nktprs); kxprs = prs(nktprs+1:nktprs+nkxprs); dc = prs(nktprs+nkxprs+1); XXspprs = prs(nktprs+nkxprs+2:end); % Extract some other stuff we'll use a lot XXstm = Xstruct.Xstim; % stimulus design matrix XXsp = Xstruct.Xsp; % spike history design matrix bsps = Xstruct.bsps; % binary spike vector M = Xstruct.Minterp; % matrix for interpolating from stimulus bins to spike train bins ihflag = Xstruct.ihflag; % flag rlen = Xstruct.rlen; % number of bins in spike train vector nsp = (sum(bsps)); % number of spikes dt = Xstruct.dtSp; % absolute bin size for spike train (in sec) % -- Make block-diag matrix containing nkt params ----- Mkt = kron(speye(nkx),reshape(ktprs,nkt,krank)); % block-diagonal matrix for kt params inds = reshape((1:nkx*krank),krank,nkx)'; inds = inds(:); Mkt = Mkt(:,inds); % turns out we need it reshaped this way % -------- Compute bilinear stim filter reponse ----------------------- dSSdx = XXstm*Mkt; % stim filtered with temporal filter ystm = dSSdx*kxprs; % -------- Compute sum of filter reponses ----------------------- if ihflag Itot = M*ystm + XXsp*XXspprs + dc; % stim-dependent + spikehist-dependent inputs else Itot = M*ystm + dc; % stim-dependent input only end % --------- Compute output of nonlinearity ------------------------ [rr,drr,ddrr] = Xstruct.nlfun(Itot); % Check for (approximately) zero-values of rr etol = 1e-100; % cutoff for small values of conditional intensity iiz = find(rr <= etol); rr(iiz) = etol; % Set value to small drr(iiz) = 0; ddrr(iiz) = 0; % Set derivs here to 0 % --------- Compute log-likelXXspood --------------------------------- Trm1 = sum(rr)*dt; % non-spike term Trm2 = -sum(log(rr(bsps))); % spike term logli = Trm1 + Trm2; % ============================================= % --------- Compute Gradient ----------------- % ============================================= if (nargout > 1) % Compute dSSdt - stim filtered with spatial kernel Mkx = kron(speye(nkt), reshape(kxprs,nkx,krank)); inds = reshape((1:nkt*krank),krank,nkt)'; inds = inds(:); Mkx = Mkx(:,inds); inds = reshape((1:nkt*nkx),nkt,nkx)'; inds = inds(:); dSSdt = XXstm(:,inds)*Mkx; % Non-spiking terms (Term 1) dqq = drr'*M; dLdkx0 = (dqq*dSSdx)'; dLdkt0 = (dqq*dSSdt)'; dLdb0 = sum(drr); if ihflag, dLdh0 = (drr'*XXsp)'; end % Non-spiking terms (Term 2) Msp = M(bsps,:); % interpolation matrix just for spike bins frac1 = drr(bsps)./rr(bsps); fracsp = frac1'*Msp; dLdkx1 = (fracsp*dSSdx)'; dLdkt1 = (fracsp*dSSdt)'; dLdb1 = sum(frac1); if ihflag, dLdh1 = (frac1'*XXsp(bsps,:))'; end % Combine Term 1 and Term 2 dLdkx = dLdkx0*dt - dLdkx1; dLdkt = dLdkt0*dt - dLdkt1; dLdk = [dLdkt; dLdkx]; dLdb = dLdb0*dt - dLdb1; if ihflag, dLdh = dLdh0*dt - dLdh1; else dLdh = []; end dL = [dLdk; dLdb; dLdh]; end % ============================================= % --------- Compute Hessian ----------------- % ============================================= if nargout > 2 % --- Non-spiking terms ----- % multiply each row of M with drr ddrrdiag = spdiags(ddrr,0,rlen,rlen); ddqqIntrp = ddrrdiag*M; ddqq = M'*ddqqIntrp; % this is MUCH faster than using bsxfun, due to sparsity! % Hx and Ht Hkt = (dSSdt'*ddqq*dSSdt)*dt; Hkx = (dSSdx'*ddqq*dSSdx)*dt; % Hxt Hktx0a= reshape((drr'*M)*XXstm,nkt,nkx); Hktx0a = kron(speye(krank),Hktx0a); Hktx = (dSSdx'*ddqq*dSSdt + Hktx0a')*dt; % Hb Hb = sum(ddrr)*dt; % Hkb sumqq = sum(ddqqIntrp,1); Hxb = (sumqq*dSSdx)'; Htb = (sumqq*dSSdt)'; Hkb = [Htb;Hxb]*dt; if ihflag % Hh Hh = (XXsp'*ddrrdiag*XXsp)*dt; % Hhk Hth = (XXsp'*ddqqIntrp)*dSSdt; Hxh = (XXsp'*ddqqIntrp)*dSSdx; Hkh = [Hth'; Hxh']*dt; % Hb Hhb = (ddrr'*XXsp)'*dt; else Hkh = []; Hhb=[]; Hh=[]; end % --- Add in spiking terms ---- frac2 = (rr(bsps).*ddrr(bsps) - drr(bsps).^2)./rr(bsps).^2; fr2diag = spdiags(frac2,0,nsp,nsp); fr2Interp = fr2diag*Msp; fr2ddqq = Msp'*fr2Interp; % Spiking terms, Hxx and Htt Hkt= Hkt - dSSdt'*fr2ddqq*dSSdt; Hkx= Hkx - dSSdx'*fr2ddqq*dSSdx; % Spiking terms, Hxt Hktx1a= reshape(frac1'*Msp*XXstm,nkt,nkx)'; Hktx1a= kron(speye(krank),Hktx1a); Hktx1b= dSSdx'*fr2ddqq*dSSdt; Hktx1= Hktx1a+Hktx1b; Hktx= Hktx - Hktx1; % Const term Hb = Hb-sum(frac2); % Const by k sumfr2 = sum(fr2Interp,1); Hxb1 = sumfr2*dSSdx; Htb1 = sumfr2*dSSdt; Hkb = Hkb - [Htb1'; Hxb1']; if ihflag XXsprrsp = fr2diag*XXsp(bsps,:); Hh= Hh - XXsp(bsps,:)'*XXsprrsp; % Const by h Hhb1 = sum(XXsprrsp,1)'; Hhb = Hhb - Hhb1; % k by h term Hth0 = XXsp(bsps,:)'*(fr2Interp*dSSdt); Hxh0 = XXsp(bsps,:)'*(fr2Interp*dSSdx); Hkh = Hkh-[Hth0';Hxh0']; end Hk = [[Hkt; Hktx] [Hktx'; Hkx]]; H = [[Hk Hkb Hkh]; [Hkb' Hb Hhb']; [Hkh' Hhb Hh]]; end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/MAPfit_GLM.m
.m
3,158
97
function [gg,neglogli,H,Xstruct,neglogp] = MAPfit_GLM(gg,Stim,Cinv,optimArgs) % [gg,neglogli,H,Xstruct,neglogp] = MAPfit_GLM(gg,C,Stim,optimArgs) % % Computes the MAP estimate for GLM params, using grad and hessians under % a zero-mean Gaussian prior with inverse covariance Cinv. % % Minimizes negative log-likelihood plus a penalty of the form % 0.5*x'*Cinv*x, where x is the parameter vector % % Assumes basis for temporal dimensions of stim filter % % Inputs: % ------- % gg = param struct % Stim = stimulus % Cinv = inverse of prior covariance matrix (gives quadratic penalty) % optimArgs = cell array of optimization params (optional) % % Outputs: % ------- % ggnew = new param struct (with MAP params); % neglogli = negative log-likelihood at MAP estimate % H = Hessian of negative log-likelihood at MAP estimate % Xstruct = structure with design matrices for spike-hist and stim terms % neglogp = negative log-posterior at MAP estimate % Set optimization parameters algopts = getFminOptsForVersion(version); if nargin > 3, opts = optimset(algopts{:}, optimArgs{:}); else, opts = optimset(algopts{:}); end % --- Create design matrix extract initial params from gg ---------------- [prs0,Xstruct] = setupfitting_GLM(gg,Stim); % --- Set log-likelihood function ---------------------------- if isequal(Xstruct.nlfun,@expfun) || isequal(Xstruct.nlfun,@exp) % loss function for exponential nonlinearity lfunc = @(prs)Loss_GLM_logli_exp(prs,Xstruct); else lfunc = @(prs)Loss_GLM_logli(prs,Xstruct); % loss function for all other nonlinearities end % set log-posterior function lpost = @(prs)(neglogpost(prs,lfunc,Cinv)); % --- minimize negative log posterior -------------------- [prsMAP,neglogp] = fminunc(lpost,prs0,opts); % find MAP estimate of params % Compute Hessian of log-likelihood, if desired if nargout > 1 [neglogli,~,H] = lfunc(prsMAP); end % Put returned vals back into param structure gg = reinsertFitPrs_GLM(gg,prsMAP,Xstruct); end %% ================================================= function [negLP,dLP,H] = neglogpost(prs,lfunc,Cinv) % Compute log-posterior by adding quadratic penalty to log-likelihood nprs = size(Cinv,1); % number of parameters in C preg = prs(1:nprs); % parameters being regularized. switch nargout case 1 % evaluate function negLP = lfunc(prs) + .5*preg'*Cinv*preg; case 2 % evaluate function and gradient [negLP,dLP] = lfunc(prs); negLP = negLP + .5*preg'*Cinv*preg; dLP(1:nprs) = dLP(1:nprs) + Cinv*preg; case 3 % evaluate function and gradient [negLP,dLP,H] = lfunc(prs); negLP = negLP + .5*preg'*Cinv*preg; dLP(1:nprs) = dLP(1:nprs) + Cinv*preg; H(1:nprs,1:nprs) = H(1:nprs,1:nprs) + Cinv; end end % %---------------------------------------------------- % Optional debugging code % %---------------------------------------------------- % % % ------ Check analytic gradients and Hessians ------- % HessCheck(lfunc,prs0,opts); % HessCheck_Elts(@Loss_GLM_logli, [1 12],prs0,opts); % tic; [lival,J,H]=lfunc(prs0); toc;
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/MLfit_GLMbi_coordascent.m
.m
3,484
113
function [gg,fval,H] = MLfit_GLMbi_coordascent(gg,Stim,maxiter,ftol,optimArgs) % [gg,fval,H] = MLfit_GLMbi_coordascent(gg,Stim,maxiter,ftol,optimArgs); % % Computes the ML estimate for GLM params, with gradient and hessians, via % coordinate ascent, for bilinear (low rank) parametrization of space-time filter. % % Inputs: % gg = param struct % Stim = stimulus % maxiter = maximum number of coordinate ascent steps % ftol = tolerance for stopping based on function improvement % optimArgs = cell array of optimization params (optional) % % Outputs: % ggnew = new param struct (with ML params); % fval = negative log-likelihood at ML estimate % H = Hessian of negative log-likelihood at ML estimate % Set maxiter if necessary if (nargin <= 2) || isempty(maxiter) maxiter = 50; % maximum number of coordinate ascent steps end if (nargin <=3) || isempty(ftol) ftol = .001; % tolerance for stopping end % Set optimization parameters if nargin < 5 optimArgs = []; end % create struct for spatial optimization ggx = gg; ggx.ktype = 'linear'; % set type to linear (from bilinear) ggx.ktbas = 1; ggx.ktbasprs = []; % remove temporal basis ggx.kt = gg.kx(:)'; ggx.k = ggx.kt; % create struct for temporal optimization ggt = gg; ggt.ktype = 'linear'; % set type to linear (from bilinear) ggt.kt = gg.kt; ggt.k = ggt.ktbas*ggt.kt; % Initialize spatial stimulus [nt,nkx] = size(Stim); krank = gg.krank; xStim = zeros(nt,krank*nkx); % initialize spatial stimulus % compute initial log-likelihood neglogli0 = neglogli_GLM(gg,Stim); % Compute logli of initial params dlogli = inf; % initialize change in logli jjiter = 0; % initialize counter while (jjiter<maxiter) && dlogli>ftol % ---- Update temporal params ----- fprintf('Iter #%d: Updating temporal params\n', jjiter); tStim = Stim*reshape(ggx.k',[],gg.krank); ggt.dc = ggx.dc; % update dc param [ggt,tneglogli] = MLfit_GLM(ggt,tStim,optimArgs); fprintf(' dlogli = %.4f\n', neglogli0-tneglogli); % Convolve stimulus with temporal filters for irank = 1:krank for icol = 1:nkx xStim(:,icol+(irank-1)*nkx) = sameconv(Stim(:,icol),ggt.k(:,irank)); end end % ---- Update spatial params ---- fprintf('Iter #%d: Updating spatial params\n', jjiter); ggx.dc = ggt.dc; % update dc param [ggx,xneglogli] = MLfit_GLM(ggx,xStim,optimArgs); fprintf(' dlogli = %.4f\n', neglogli0-xneglogli); % Update iters jjiter = jjiter+1; % counter dlogli = neglogli0-xneglogli; % change in log-likelihood neglogli0 = xneglogli; end fprintf('\nFinished coordinate ascent: %d iterations (dlogli=%.6f)\n',jjiter,dlogli); % Compute conditional Hessians, if desired if nargout > 2 % Compute Hessian for time components tStim = Stim*(ggx.k'); ggt.dc = ggx.dc; % update dc param [ggt,~,Ht] = MLfit_GLM(ggt,tStim,optimArgs); % Compute Hessian for space components for irank = 1:krank for icol = 1:nkx xStim(:,icol+(irank-1)*nkx) = sameconv(Stim(:,icol),ggt.k(:,irank)); end end ggx.dc = ggt.dc; % update dc param [ggx,neglogli0,Hx] = MLfit_GLM(ggx,xStim,optimArgs); H = {Ht,Hx}; % Hessians for t and x components end % Update params of bilinear model gg.dc = ggx.dc; gg.kt = ggt.kt; gg.kx = reshape(ggx.k',[],gg.krank); gg.k = (gg.ktbas*gg.kt)*gg.kx'; fval = neglogli0; % value of loglikelihood
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/initfit_mask.m
.m
1,147
38
function bmask = initfit_mask(mask,dtSp,rlen) % bmask = initfit_mask(mask,dtSp,rlen) % % Compute binary mask for computing log-likelihood % % Input: % ------ % mask [n x 2]: list of intervals to use for computing log-likelihood % (ignore time bins outside these intervals) % dtSp [1 x 1]: bin size for spike trains and conditional intensity % rlen [1 x 1]: length of total conditional intensity vector % % Output: % ------- % bmask [rlen x 1]: binary vector where (1 = use for LL), (0 = ignore). if isempty(mask) % No mask bmask = true(rlen,1); % keep all indices else % convert mask to discrete time bins mask = round(mask/dtSp); % Check that mask doesn't extend beyond stimulus window if max(mask(:)) > rlen warning('GLM:mask', 'Mask is too long for data segment: truncating...'); mask(:,2) = max(mask(:,2),rlen); % set maximum mask value to stim length mask(diff(mask')'<0,:) = []; % remove rows with negative intervals end % Generate mask bmask = false(rlen,1); for j = 1:size(mask,1) bmask(mask(j,1):mask(j,2)) = true; end end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/makeFittingStruct_GLM.m
.m
3,351
78
function gg = makeFittingStruct_GLM(dtStim,dtSp,klength,nkbasis,k0,nhbasis,lasthpeak) % gg = makeFittingStruct_GLM(dtStim,dtSp,klength,nkbasis,k0,nhbasis,lasthpeak) % % Initialize parameter structure for fitting GLM, % with normal parametrization of stim kernel % % Inputs: % dtStim = bin size of stimulus (in s) % dtSp = bin size for spike train (in s) % klength = temporal length of stimulus filter, in # of bins (optional) % nkbasis = # temporal basis vectors for stim filter (optional) % nhbasis = # temporal basis vectors for spike hist filter h (optional) % lasthpeak = time of peak of last h basis vector, in s (optional) % k0 = initial estimate of stimulus filter (optional) % % Outputs: % gg = GLM-fitting param structure % ---- Set up fitting structure ------------------------------- gg.k = []; % Actual stim filter k gg.ih = []; % Actual post-spike filter ih gg.dc = 0; % "dc" or constant input (determines baseline spike rate) gg.ihw = []; % h filter weights gg.iht = []; % h filter time points gg.ihbas = []; % basis for h filter gg.ihbasprs = []; % parameters for basis for h-filter gg.kt = []; % basis weights for stimulus filter k gg.ktbas = []; % temporal basis for stimulus filter k gg.ktbasprs = []; % parameters for basis for k-filter gg.nlfun = @expfun; % default nonlinearity: exponential gg.sps = []; % spike times (in s) gg.mask = []; % list of intervals to ignore when computing likelihood gg.dtStim = dtStim; % time bin size for stimulus gg.dtSp = dtSp; % time bin for spike train gg.ihw2 = []; % weights for coupling filters gg.ihbas2 = []; % basis for coupling filters gg.ihbasprs2 = []; % parameters for coupling filter basis gg.sps2 = []; % spike times of coupled neurons gg.couplednums = []; % numbers of coupled cells % % ----- Set up temporal basis for stimulus kernel ----------- if nargin > 2 assert((klength>nkbasis), 'klength should be bigger than number of temporal basis vectors'); ktbasprs.neye = 0; % number of "identity" basis vectors ktbasprs.ncos = nkbasis; % Number of raised-cosine vectors to use ktbasprs.kpeaks = [0 klength*(1 - 1.5/nkbasis)]; % Position of 1st and last bump ktbasprs.b = 10; % Offset for nonlinear scaling (larger -> more linear) [~,ktbasis] = makeBasis_StimKernel(ktbasprs,klength); gg.ktbas = ktbasis; gg.ktbasprs = ktbasprs; end if (nargin > 4) && (~isempty(k0)) % initialize k filter in this basis gg.kt = (gg.ktbas'*gg.ktbas)\(gg.ktbas'*k0); gg.k = gg.ktbas*gg.kt; end % ----- Set up basis for post-spike filter ----------------------- if (nargin > 5) && (nhbasis>0) ihbasprs.ncols = nhbasis; % Number of basis vectors for post-spike kernel ihbasprs.hpeaks = [dtSp lasthpeak]; % Peak location for first and last vectors ihbasprs.b = lasthpeak/5; % How nonlinear to make spacings (rough heuristic) ihbasprs.absref = []; % absolute refractory period (optional) [iht,ihbas] = makeBasis_PostSpike(ihbasprs,dtSp); gg.iht = iht; gg.ihbas = ihbas; gg.ihbasprs = ihbasprs; gg.ihw = zeros(size(ihbas,2),1); gg.ih = gg.ihbas*gg.ihw; % Set ih to be the current value of ihbas*ihw end % specify type of parametrization of filter ('linear' vs. 'bilinear') gg.ktype = 'linear';
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/initfit_sphistDesignMat.m
.m
1,390
48
function Xstruct = initfit_sphistDesignMat(gg,Xstruct) % Xstruct = initfit_sphistDesignMat(gg,Xstruct) % % Sets parameters relating to optimization of % spike-filter terms in the point process GLM and inserts them into design % matrix structure 'Xstruct' % Determine # of parameters for self-coupling filter if ~isempty(gg.ihw), nh=size(gg.ihbas,2); else nh=0; end % Determine # of params for coupling filters if ~isempty(gg.ihw2), nh2 = size(gg.ihbas2,2); else nh2=0; end % Determine # of coupled neurons nCoupled = length(gg.couplednums); % Number of coupled cells % Set flag for existence of coupling filters if (nh+nh2 == 0), Xstruct.ihflag = false; else Xstruct.ihflag = 1; end % Vector of binary (0 or 1) spike counts Xstruct.bsps = sparse(gg.sps>0); % ---- Create Spike-history Design Matrix ------------------------ Xsp = zeros(Xstruct.rlen,nh+nh2*nCoupled); % allocate % Design matrix for self-coupling filter if nh>0 Xsp(:,1:nh) = spikefilt(full(double(gg.sps>0)),gg.ihbas); end % Design matrix for cross-coupling filters for jcpl = 1:nCoupled inds = nh+nh2*(jcpl-1)+1:nh+nh2*jcpl; % column indices Xsp(:,inds) = spikefilt(full(double(gg.sps2(:,jcpl))),gg.ihbas2); % insert into design matrix end % ---- Set fields of Xstruct ------------------------------------- Xstruct.nh = nh; Xstruct.nh2 = nh2; Xstruct.nCoupled = nCoupled; Xstruct.Xsp = Xsp;
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/setupfitting_GLM.m
.m
1,529
45
function [prs0,Xstruct] = setupfitting_GLM(gg, Stim) % [prs0,Xstruct] = setupfitting_GLM(gg, Stim) % % Set initial parameters and build design matrix structure for GLM fitting % % Inputs: % gg = glm param structure % Stim = stimulus (time along columns, other dims along rows) % maxsize = maximum # floats to store in design matrix % (Affects # of chunks used to compute likelihood) % Output: % prs0 = initial parameters extracted from gg % Xstruct = struct with design matrix for stim and spike-history terms % Initialize optimization param structure % ---- Create struct and make stimulus design matrix --------------------- if strcmp(gg.ktype, 'linear') % standard GLM Xstruct = initfit_stimDesignMat(gg,Stim); % create design matrix structure % extract parameter vector prs0 = [gg.kt(:); gg.dc; gg.ihw(:); gg.ihw2(:)]; elseif strcmp(gg.ktype, 'bilinear') % bilinearly-parametrized stim filter Xstruct = initfit_stimDesignMat_bi(gg,Stim); % create design matrix structure % extract parameter vector prs0 = [gg.kt(:); gg.kx(:); gg.dc; gg.ihw(:); gg.ihw2(:)]; else error('unknown filter type (allowed types are ''linear'' or ''bilinear'')'); end % ---- Make spike-history design matrix ----------------------------------- Xstruct = initfit_sphistDesignMat(gg,Xstruct); % set nonlinearity Xstruct.nlfun = gg.nlfun; % compute mask (time bins to use for likelihood) Xstruct.bmask = initfit_mask(gg.mask,Xstruct.dtSp,Xstruct.rlen);
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/Loss_GLM_logli.m
.m
4,129
122
function [logli, dL, H] = Loss_GLM_logli(prs,Xstruct) % [logli, dL, H] = Loss_GLM_logli(prs,Xstruct) % % Compute negative log-likelihood of data undr the GLM model % (with standard linear parametrization of stimulus kernel); % % Uses arbitrary nonlinearity 'nlfun' instead of exponential % % Inputs: % prs = [kprs - weights for stimulus kernel % dc - dc current injection % ihprs - weights on post-spike current] % % Outputs: % logli = negative log likelihood of spike train % dL = gradient with respect to prs % H = hessian % Unpack GLM prs; nktot = Xstruct.nkx*Xstruct.nkt; % total # params for k kprs = prs(1:nktot); dc = prs(nktot+1); ihprs = prs(nktot+2:end); % Extract some other stuff we'll use a lot XXstm = Xstruct.Xstim; % stimulus design matrix XXsp = Xstruct.Xsp; % spike history design matrix bsps = Xstruct.bsps; % binary spike vector M = Xstruct.Minterp; % matrix for interpolating from stimulus bins to spike train bins ihflag = Xstruct.ihflag; % flag rlen = Xstruct.rlen; % number of bins in spike train vector nsp = (sum(bsps)); % number of spikes dt = Xstruct.dtSp; % absolute bin size for spike train (in sec) % -------- Compute sum of filter reponses ----------------------- if Xstruct.ihflag Itot = M*(XXstm*kprs) + XXsp*ihprs + dc; % stim-dependent + spikehist-dependent inputs else Itot = M*(XXstm*kprs) + dc; % stim-dependent input only end % --------- Compute output of nonlinearity ------------------------ [rr,drr,ddrr] = Xstruct.nlfun(Itot); % --------- Compute log-likelihood --------------------------------- Trm1 = sum(rr)*dt; % non-spike term Trm2 = -sum(log(rr(bsps))); % spike term logli = Trm1 + Trm2; % --------- Compute Gradient ----------------- if (nargout > 1) % Non-spiking terms (Term 1) dLdk0 = (drr'*M*XXstm)'; dLdb0 = sum(drr); if ihflag, dLdh0 = (drr'*XXsp)'; end % Spiking terms (Term 2) Msp = M(bsps,:); % interpolation matrix just for spike bins frac1 = drr(bsps)./rr(bsps); dLdk1 = ((frac1'*Msp)*XXstm)'; dLdb1 = sum(frac1); if ihflag, dLdh1 = (frac1'*XXsp(bsps,:))'; end % Combine Term 1 and Term 2 dLdk = dLdk0*dt - dLdk1; dLdb = dLdb0*dt - dLdb1; if ihflag, dLdh = dLdh0*dt - dLdh1; else dLdh = []; end dL = [dLdk; dLdb; dLdh]; end % --------- Compute Hessian ----------------- if nargout > 2 % --- Non-spiking terms ----- % multiply each row of M with drr ddrrdiag = spdiags(ddrr,0,rlen,rlen); ddqqIntrp = ddrrdiag*M; % this is MUCH faster than using bsxfun, due to sparsity! % k and b terms Hk = (XXstm'*(M'*ddqqIntrp)*XXstm)*dt; % Hkk (k filter) Hb = sum(ddrr)*dt; % Hbb (constant b) Hkb = (sum(ddqqIntrp,1)*XXstm)'*dt; % Hkb (cross-term) if ihflag % h terms Hh = (XXsp'*(bsxfun(@times,XXsp,ddrr)))*dt; % Hh (h filter) % (here bsxfun is faster than diagonal multiplication) Hkh = ((XXsp'*ddqqIntrp)*XXstm*dt)'; % Hhk (cross-term) Hhb = (ddrr'*XXsp)'*dt; % Hhb (cross-term) else Hkh=[]; Hhb=[]; Hh=[]; end % --- Add in spiking terms ---- frac2 = (rr(bsps).*ddrr(bsps) - drr(bsps).^2)./rr(bsps).^2; % needed weights from derivation of Hessian fr2Interp = spdiags(frac2,0,nsp,nsp)*Msp; % rows of Msp re-weighted by these weights % Spiking terms, k and b Hk= Hk - XXstm'*(Msp'*fr2Interp)*XXstm; % Hkk (k filter) Hb = Hb-sum(frac2); % Hbb (constant b) Hb1 = sum(fr2Interp,1)*XXstm; % Spiking term, k and const Hkb = Hkb - Hb1'; if Xstruct.ihflag XXrrsp = bsxfun(@times,XXsp(bsps,:),frac2); Hh= Hh - XXsp(bsps,:)'*XXrrsp; % Const by h Hhb1 = sum(XXrrsp,1)'; Hhb = Hhb - Hhb1; % k by h term Hkh0 = XXsp(bsps,:)'*(fr2Interp*XXstm); Hkh = Hkh-Hkh0'; end H = [[Hk Hkb Hkh]; [Hkb' Hb Hhb']; [Hkh' Hhb Hh]]; end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/initfit_stimDesignMat.m
.m
1,536
36
function Xstruct = initfit_stimDesignMat(gg,Stim) % Xstruct = initfit_stimDesignMat(gg,Stim) % % Initialize parameters relating to stimulus design matrix % ---- Set up filter and stim processing params ------------------- nkx = size(gg.k,2); % number stim pixels (# x params) nkt = size(gg.ktbas,2); % # time params per stim pixel (# t params) ncols = nkx*nkt; % total number of columns in stim design matrix [slen,swid] = size(Stim); % size of stimulus upsampfactor = (gg.dtStim/gg.dtSp); % number of times by which spikes more finely sampled than stim rlen = slen*upsampfactor; % length of spike train vector % ---- Check size of filter and width of stimulus ---------- assert(nkx == swid,'Mismatch between stim width and kernel width'); % ---- Convolve stimulus with spatial and temporal bases ----- Xstruct.Xstim = zeros(slen,ncols); for i = 1:nkx for j = 1:nkt Xstruct.Xstim(:,(i-1)*nkt+j) = sameconv(Stim(:,i),gg.ktbas(:,j)); end end % ---- Set fields of Xstruct ------------------------------------- Xstruct.nkx = nkx; % # stimulus spatial stimulus pixels Xstruct.nkt = nkt; % # time bins in stim filter Xstruct.slen = slen; % Total stimulus length (coarse bins) Xstruct.rlen = rlen; % Total spike-train bins (fine bins) Xstruct.upsampfactor = upsampfactor; % rlen / slen Xstruct.Minterp = kron(speye(slen),ones(upsampfactor,1)); % rlen x slen matrix for upsampling and downsampling Xstruct.dtStim = gg.dtStim; % time bin size for stimulus Xstruct.dtSp = gg.dtSp; % time bins size for spike train
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/makeFittingStruct_GLMbi.m
.m
1,806
53
function gg = makeFittingStruct_GLMbi(krank,varargin) % gg = makeFittingStruct_GLM(krank,dtStim,dtSp,klength,nkbasis,k0,nhbasis,lasthpeak) % % Initialize parameter struct for fitting generalized linear model (GLM), % with bilinear (i.e., low-rank) parametrization stimulus filter % % Inputs: % krank = rank of stim filter % dtStim = bin size of stimulus (in s) % dtSp = bin size for spike train (in s) % klength = temporal length of stimulus filter, in # of bins (optional) % nkbasis = # temporal basis vectors for stim filter (optional) % nhbasis = # temporal basis vectors for spike hist filter h (optional) % lasthpeak = time of peak of last h basis vector, in s (optional) % k0 = initial estimate of stimulus filter (optional) % % Outputs: % gg = GLM-fitting param structure % ==================================================================== % Set up structure gg = makeFittingStruct_GLM(varargin{:}); % Set additional fields needed by bilinear GLM gg.kx = []; gg.kxbas=[]; gg.kbasprs = []; gg.krank = krank; gg.ktype = 'bilinear'; % if initial filter passed in, use svd to set up initial K params (bilinearly parametrized) if (length(varargin)>4) && (~isempty(varargin{3})) k0 = varargin{5}; [u,s,v] = svd(k0); mux = mean(k0)'; nkt = size(gg.ktbas,2); nkx = size(k0,2); kt = zeros(nkt,krank); kx = zeros(nkx,krank); for j = 1:krank; if v(:,j)'*mux < 0 % Flip sign if shape looks opposite to k0 u(:,j) = -u(:,j); v(:,j) = -v(:,j); end kt(:,j) = (gg.ktbas'*gg.ktbas)\(gg.ktbas'*(u(:,j)*s(j,j))); kx(:,j) = v(:,j); end gg.kxbas = speye(nkx); gg.kt = kt; gg.kx = kx; gg.k = (gg.ktbas*gg.kt)*(gg.kxbas*gg.kx)'; end
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/MLfit_GLM.m
.m
1,912
53
function [gg,neglogli,H,Xstruct] = MLfit_GLM(gg,Stim,optimArgs) % [gg,neglogli,H,Xstruct] = MLfit_GLM(gg,Stim,optimArgs) % % Computes the ML estimate for GLM params, using grad and hessians. % Assumes basis for temporal dimensions of stim filter % % Inputs: % gg = param struct % Stim = stimulus % optimArgs = cell array of optimization params (optional) % % Outputs: % ggnew = new param struct (with ML params); % neglogli = negative log-likelihood at ML estimate % H = Hessian of negative log-likelihood at ML estimate % Xstruct = structure with design matrices for spike-hist and stim terms % Set optimization parameters algopts = getFminOptsForVersion(version); if nargin > 2, opts = optimset(algopts{:}, optimArgs{:}); else, opts = optimset(algopts{:}); end % --- Create design matrix extract initial params from gg ---------------- [prs0,Xstruct] = setupfitting_GLM(gg,Stim); % --- Set loss function -------------------------------------------------- if isequal(Xstruct.nlfun,@expfun) || isequal(Xstruct.nlfun,@exp) lfunc = @(prs)Loss_GLM_logli_exp(prs,Xstruct); % loss function for exponential nonlinearity else lfunc = @(prs)Loss_GLM_logli(prs,Xstruct); % loss function for all other nonlinearities end % --- minimize negative log likelihood -------------------- [prsML,neglogli] = fminunc(lfunc,prs0,opts); % find ML estimate of params % Compute Hessian if desired if nargout > 2 [neglogli,~,H] = Loss_GLM_logli(prsML,Xstruct); end % Put returned vals back into param structure ------ gg = reinsertFitPrs_GLM(gg,prsML,Xstruct); % %---------------------------------------------------- % Optional debugging code % %---------------------------------------------------- % % % ------ Check analytic gradients and Hessians ------- % HessCheck(lfunc,prs0,opts); % HessCheck_Elts(@Loss_GLM_logli, [1 12],prs0,opts); % tic; [lival,J,H]=lfunc(prs0); toc;
MATLAB
2D
pillowlab/GLMspiketools
glmtools_fitting/Loss_GLM_logli_exp.m
.m
3,269
100
function [logli, dL, H] = Loss_GLM_logli_exp(prs,Xstruct) % [neglogli, dL, H] = Loss_GLM_logli_exp(prs) % % Compute negative log-likelihood of data undr the GLM model with % exponential nonlinearity (with standard linear parametrization of stim filter) % % Inputs: % prs = [kprs - weights for stimulus kernel % dc - dc current injection % ihprs - weights on post-spike current]; % Outputs: % logli = negative log likelihood of spike train % dL = gradient with respect to prs % H = hessian % Extract some vals from Xstruct (Opt Prs); nktot = Xstruct.nkx*Xstruct.nkt; % total # params for k dt = Xstruct.dtSp; % absolute bin size for spike train (in sec) % Unpack GLM prs; kprs = prs(1:nktot); dc = prs(nktot+1); ihprs = prs(nktot+2:end); % Extract some other stuff we'll use a lot XXstm = Xstruct.Xstim; % stimulus design matrix XXsp = Xstruct.Xsp; % spike history design matrix bsps = Xstruct.bsps; % binary spike vector M = Xstruct.Minterp; % matrix for interpolating from stimulus bins to spike train bins ihflag = Xstruct.ihflag; % flag rlen = Xstruct.rlen; % number of bins in spike train vector nsp = sum(bsps); % number of spikes % -------- Compute sum of filter reponses ----------------------- if Xstruct.ihflag Itot = M*(XXstm*kprs) + XXsp*ihprs + dc; % stim-dependent + spikehist-dependent inputs else Itot = M*(XXstm*kprs) + dc; % stim-dependent input only end % --------- Compute output of nonlinearity ------------------------ rr = exp(Itot); % --------- Compute log-likelihood --------------------------------- Trm1 = sum(rr)*dt; % non-spike term Trm2 = -sum(Itot(bsps)); % spike term logli = Trm1 + Trm2; % --------- Compute Gradient ----------------- if (nargout > 1) % Non-spiking terms (Term 1) dLdk0 = (rr'*M*XXstm)'; dLdb0 = sum(rr); if ihflag, dLdh0 = (rr'*XXsp)'; end % Spiking terms (Term 2) Msp = M(bsps,:); % interpolation matrix just for spike bins dLdk1 = (sum(Msp*XXstm))'; dLdb1 = nsp; if ihflag, dLdh1 = sum(XXsp(bsps,:),1)'; end % Combine terms dLdk = dLdk0*dt - dLdk1; dLdb = dLdb0*dt - dLdb1; if ihflag, dLdh = dLdh0*dt - dLdh1; else dLdh = []; end dL = [dLdk; dLdb; dLdh]; end % --------- Compute Hessian ----------------- if nargout > 2 % --- Non-spiking terms ----- % multiply each row of M with drr ddrrdiag = spdiags(rr,0,rlen,rlen); ddqqIntrp = ddrrdiag*M; % this is MUCH faster than using bsxfun, due to sparsity! % k and b terms Hk = (XXstm'*(M'*ddqqIntrp)*XXstm)*dt; % Hkk (k filter) Hb = dLdb0*dt; % Hbb (constant b) Hkb = (sum(ddqqIntrp,1)*XXstm)'*dt; % Hkb (cross-term) if ihflag % h terms Hh = (XXsp'*(bsxfun(@times,XXsp,rr)))*dt; % Hh (h filter) % (here bsxfun is faster than diagonal multiplication) Hkh = ((XXsp'*ddqqIntrp)*XXstm*dt)'; % Hhk (cross-term) Hhb = (rr'*XXsp)'*dt; % Hhb (cross-term) else Hkh=[]; Hhb=[]; Hh=[]; end H = [[Hk Hkb Hkh]; [Hkb' Hb Hhb']; [Hkh' Hhb Hh]]; end
MATLAB
2D
pillowlab/GLMspiketools
unittests/unit_condIntensityConsistency_cpl.m
.m
3,361
85
% Unit test to check the consistency of conditional intensity computed % during simulation and during fitting, using spikes from generated from a % simulation of coupled neurons in demo3 % 1. First, run relevant section of demo3 dtStim = .01; % Bin size for simulating model & computing likelihood (in units of stimulus frames) dtSp = .001; % Bin size for simulating model & computing likelihood (must evenly divide dtStim); nkt = 20; % Number of time bins in stimulus filter k ggsim1 = makeSimStruct_GLM(nkt,dtStim,dtSp); % Create GLM struct with default params ggsim2 = makeSimStruct_GLM(nkt,dtStim,dtSp); % Create GLM struct with default params % Change second neuron's stimulus filter [ktbas,ktbasis] = makeBasis_StimKernel(ggsim2.ktbasprs,nkt); ggsim2.k = ktbasis*[0 0 .1 .25 .5 .5 .25 -.25 -.5 -.25]'; % more delayed filter ggsim = makeSimStruct_GLMcpl(ggsim1,ggsim2); % Make some coupling kernels [iht,ihbas,ihbasis] = makeBasis_PostSpike(ggsim.ihbasprs,dtSp); hhcpl = ihbasis*[.25;.25;.125;0;0]; hhcpl(:,2) = ihbasis*[-2;-1;0;.25;.25]; ggsim.ih(:,2,1) = hhcpl(:,2); % 2nd cell coupling to first ggsim.ih(:,1,2) = hhcpl(:,1); % 1st cell coupling to second % Run short simulation for visualization purposes slen = 50; % Stimulus length (frames) & width (# pixels) swid = size(ggsim.k,2); % stimulus width Stim = 2*randn(slen,swid); % Gaussian white noise stimulus [tsp,sps,Itot,Istm] = simGLM(ggsim, Stim); % Simulate GLM response Isp = Itot-Istm; % net spike-history output %% 2. compute conditional intensity again during log-likelihood evaluation % Make param object with "true" params ggTrue = makeFittingStruct_GLM(dtStim,dtSp); ggTrue.k = ggsim.k(:,:,1); % % insert stimulus filter ggTrue.dc = ggsim.dc(1); % insert dc value ggTrue.ih = ggsim.ih(:,:,1); % insert spike-history filter ggTrue.sps = sps(:,1); % spike times ggTrue.sps2 = sps(:,2); % spike times of coupled neuron ggTrue.couplednums = 2; % number of cell coupled to this one (for clarity) [logliTrue2,rrT2,tt2,Itot2,Istm2,Ih2] = neglogli_GLM(ggTrue,Stim); % ---- Unit tests ------- % On total log-conditional intensity assert(max(abs(Itot(:,1)-Itot2))<1e-8,'Unit failed: log-conditional intensity consistency'); % On just the stimulus component of the conditional intensity assert(max(abs(Istm(:,1)-Istm2))<1e-8,'Unit failed: spike-history filter output consistency'); %% 3. Do same for other neuron % Make param object with "true" params ggTrue = makeFittingStruct_GLM(dtStim,dtSp); ggTrue.k = ggsim.k(:,:,2); % % insert stimulus filter ggTrue.dc = ggsim.dc(2); % insert dc value ggTrue.ih = fliplr(ggsim.ih(:,:,2)); % insert spike-history filter ggTrue.sps = sps(:,2); % spike times ggTrue.sps2 = sps(:,1); % spike times of coupled neuron ggTrue.couplednums = 1; % number of cell coupled to this one (for clarity) [logliTrue3,rrT3,tt3,Itot3,Istm3,Ih3] = neglogli_GLM(ggTrue,Stim); % ---- Unit tests ------- % On total log-conditional intensity assert(max(abs(Itot(:,2)-Itot3))<1e-8,'Unit failed: log-conditional intensity consistency'); % On just the stimulus component of the conditional intensity assert(max(abs(Istm(:,2)-Istm3))<1e-8,'Unit failed: spike-history filter output consistency'); % =================================== % Report if passed % =================================== fprintf('Unit passed: condIntensityConsistency_cpl\n');
MATLAB
2D
pillowlab/GLMspiketools
unittests/unit_condIntensityConsistency.m
.m
1,954
51
% Unit testing code to ensure that conditional intensity is computed the % same way during simulation as during fitting %% 1. Set parameters and display for a GLM % ============================== dtStim = .01; % Bin size for stimulus (in seconds). (Equiv to 100Hz frame rate) dtSp = .001; % Bin size for simulating model & computing likelihood (must evenly divide dtStim); nkt = 30; % Number of time bins in stimulus filter k ttk = dtStim*(-nkt+1:0)'; % time relative to spike of stim filter taps ggsim = makeSimStruct_GLM(nkt,dtStim,dtSp); % Create GLM structure with default params %% 2. Simulate model and compute conditional intensity ==================== slen = 5000; % Stimulus length (frames); More samples gives better fit swid = 1; Stim = randn(slen,swid); % Run model on long, binary stimulus [tsp,sps,Itot,Istm] = simGLM(ggsim,Stim); % run model and return conditional intensity nsp = length(tsp); %% 3. Make param object with "true" params ================================ ggTrue = makeFittingStruct_GLM(dtStim,dtSp); ggTrue.k = ggsim.k; % insert stimulus filter ggTrue.dc = ggsim.dc; % insert dc value ggTrue.ih = ggsim.ih; % insert spike-history filter ggTrue.sps = sps; % spike times ggTrue.mask = []; %% 4. Compute intensity again (during computation of log-likelihood) ====== % (if desired, compare rrT computed here with vmem computed above). % To make it fail if you want to: % Stim(1) = Stim(1)+.1; [logliTrue, rrT,tt,Itot2,Istm2,Ih2] = neglogli_GLM(ggTrue,Stim); %% 5. Unit tests % On total log-conditional intensity assert(max(abs(Itot-Itot2))<1e-8,'Unit failed: log-conditional intensity consistency'); % On just the stimulus component of the conditional intensity assert(max(abs(Istm-Istm2))<1e-8,'Unit failed: spike-history filter output consistency'); % =================================== % Report if passed % =================================== fprintf('Unit passed: condIntensityConsistency\n');
MATLAB
2D
MRSRL/mridata-recon
recon_2d_fse.py
.py
10,357
268
"""Basic recon for 2D FSE datasets on mridata.org.""" import numpy as np import os import ismrmrd import argparse import imageio from tqdm import tqdm from fileio import cfl from mrirecon import fse from mrirecon import fftc def isrmrmd_user_param_to_dict(header): """ Store ISMRMRD header user parameters in a dictionary. Parameter --------- header : ismrmrd.xsd.ismrmrdHeader ISMRMRD header object Returns ------- dict Dictionary containing custom user parameters """ user_dict = {} user_long = list(header.userParameters.userParameterLong) user_double = list(header.userParameters.userParameterDouble) user_string = list(header.userParameters.userParameterString) user_base64 = list(header.userParameters.userParameterBase64) for entry in user_long + user_double + user_string + user_base64: user_dict[entry.name] = entry.value_ return user_dict def load_ismrmrd_to_np(file_name, verbose=False): """ Load data from an ISMRMRD file to a numpy array. Raw data from the ISMRMRD file is loaded into a numpy array. If the ISMRMRD file includes the array 'rec_std' that contains the standard deviation of the noise, this information is used to pre-whiten the k-space data. If applicable, a basic phase correction is performed on the loaded k-space data. Parameters ---------- file_name : str Name of ISMRMRD file verbose : bool, optional Turn on/off verbose print out Returns ------- np.array k-space data in an np.array of dimensions [phase, echo, slice, coils, kz, ky, kx] ismrmrd.xsd.ismrmrdHeader ISMRMRD header object """ dataset = ismrmrd.Dataset(file_name, create_if_needed=False) header = ismrmrd.xsd.CreateFromDocument(dataset.read_xml_header()) param_dict = isrmrmd_user_param_to_dict(header) num_kx = header.encoding[0].encodedSpace.matrixSize.x num_ky = header.encoding[0].encodingLimits.kspace_encoding_step_1.maximum + 1 num_kz = header.encoding[0].encodingLimits.kspace_encoding_step_2.maximum + 1 num_channels = header.acquisitionSystemInformation.receiverChannels num_slices = header.encoding[0].encodingLimits.slice.maximum + 1 num_echoes = header.encoding[0].encodingLimits.contrast.maximum + 1 num_phases = header.encoding[0].encodingLimits.phase.maximum + 1 num_segments = header.encoding[0].encodingLimits.segment.maximum + 1 is_fse_with_calib = num_segments > 1 chop_y = 1 - int(param_dict.get('ChopY', 1)) chop_z = 1 - int(param_dict.get('ChopZ', 1)) try: rec_std = dataset.read_array('rec_std', 0) rec_weight = 1.0 / (rec_std ** 2) rec_weight = np.sqrt(rec_weight / np.sum(rec_weight)) except Exception: rec_weight = np.ones(num_channels) opt_mat = np.diag(rec_weight) if verbose: print("Data dims: (%d, %d, %d, %d, %d, %d, %d)" % (num_kx, num_ky, num_kz, num_channels, num_slices, num_echoes, num_phases)) kspace = np.zeros([num_phases, num_echoes, num_slices, num_channels, num_kz, num_ky, num_kx], dtype=np.complex64) if is_fse_with_calib: echo_train = np.zeros([num_phases, num_echoes, num_slices, 1, num_kz, num_ky, 1], dtype=np.uint) kspace_fse_cal = np.zeros([num_phases, num_echoes, num_slices, num_channels, num_segments, num_kx], dtype=np.complex64) echo_train_fse_cal = np.zeros([num_phases, num_echoes, num_slices, 1, num_segments, 1], dtype=np.uint) max_slice = 0 wrap = lambda x: x if verbose: print("Loading data...") wrap = tqdm try: num_acq = dataset.number_of_acquisitions() except: print("Unable to determine number of acquisitions! Empty?") return for i in wrap(range(num_acq)): acq = dataset.read_acquisition(i) i_ky = acq.idx.kspace_encode_step_1 # pylint: disable=E1101 i_kz = acq.idx.kspace_encode_step_2 # pylint: disable=E1101 i_echo = acq.idx.contrast # pylint: disable=E1101 i_phase = acq.idx.phase # pylint: disable=E1101 i_slice = acq.idx.slice # pylint: disable=E1101 if i_slice > max_slice: max_slice = i_slice sign = (-1) ** (i_ky * chop_y + i_kz * chop_z) data = np.matmul(opt_mat.T, acq.data) * sign if i_kz < num_kz: i_segment = acq.idx.segment # pylint: disable=E1101 if i_ky < num_ky: kspace[i_phase, i_echo, i_slice, :, i_kz, i_ky, :] = data if is_fse_with_calib: echo_train[i_phase, i_echo, i_slice, 0, i_kz, i_ky, 0] = i_segment elif is_fse_with_calib: kspace_fse_cal[i_phase, i_echo, i_slice, :, i_ky - num_ky, :] = data echo_train_fse_cal[i_phase, i_echo, i_slice, 0, i_ky - num_ky, 0] = i_segment dataset.close() max_slice += 1 if num_slices != max_slice: if verbose: print("Actual number of slices different: %d/%d" % (max_slice, num_slices)) kspace = kspace[:, :, :max_slice, :, :, :, :] if is_fse_with_calib: echo_train = echo_train[:, :, :max_slice, :, :, :, :] kspace_fse_cal = kspace_fse_cal[:, :, :max_slice, :, :, :] echo_train_fse_cal = echo_train_fse_cal[:, :, :max_slice, :, :, :] if is_fse_with_calib: if verbose: print("FSE phase correction...") if 0: print("writing files for debugging...") cfl.write("kspace", kspace) cfl.write("echo_train", echo_train) cfl.write("kspace_fse_cal", kspace_fse_cal) cfl.write("echo_train_fse_cal", echo_train_fse_cal) kspace_cor = fse.phase_correction(kspace, echo_train, kspace_fse_cal, echo_train_fse_cal) # for debugging if 0: cfl.write("kspace_orig" , kspace) kspace = kspace_cor return kspace, header def transform(kspace, header, verbose=False): """ Transform kspace data to image domain. If needed, cropping is performed based on information in the header. Parameters ---------- kspace : np.array Data in k-space [phase, echo, slice, coils, z, ky, kx] header : ismrmrd.xsd.ismrmrdHeader ISMRMRD header object verbose : bool, optional Turn on/off print outs Returns ------- np.array Data in image domain [phase, echo, slice, coils, z, y, x] """ image = fftc.ifft2c(kspace) num_kx = header.encoding[0].encodedSpace.matrixSize.x num_ky = header.encoding[0].encodingLimits.kspace_encoding_step_1.maximum + 1 num_x = header.encoding[0].reconSpace.matrixSize.x num_y = header.encoding[0].reconSpace.matrixSize.y if num_x < num_kx: if verbose: print("Cropping data in x (%d to %d)..." % (num_kx, num_x)) x0 = (num_kx - num_x) // 2 x1 = x0 + num_x image = image[:, :, :, :, :, :, x0:x1] if num_y < num_ky: if verbose: print("Cropping data in y (%d to %d)..." % (num_ky, num_y)) y0 = (num_ky - num_y) // 2 y1 = y0 + num_y image = image[:, :, :, :, :, y0:y1, :] return image def dataset_to_cfl(dir_out, file_name, suffix="", file_png=None, verbose=False): """ Convert ISMRMRD to CFL files in specified directory. Parameters ---------- dir_out : str Output directory to write CFL files file_name : str Name of ISMRMRD file suffix : str, optional Suffix to attach to output file names file_png : str, optional If not None, a png file will be written out verbose : bool, optional Turn on/off verbose print outs """ kspace, header = load_ismrmrd_to_np(file_name, verbose=verbose) if verbose: print("Transforming k-space data to image domain...") image = transform(kspace, header, verbose=verbose) num_phases = kspace.shape[0] num_echoes = kspace.shape[1] if verbose: print("Writing files...") for i_phase in range(num_phases): for i_echo in range(num_echoes): suffix_i = suffix if num_echoes > 1: suffix_i = "_" + ("echo%02d" % i_echo) + suffix_i if num_phases > 1: suffix_i = "_" + ("phase%02d" % i_phase) + suffix_i cfl.write(os.path.join(dir_out, "kspace" + suffix_i), kspace[i_phase, i_echo, :, :, :, :, :]) cfl.write(os.path.join(dir_out, "image" + suffix_i), image[i_phase, i_echo, :, :, :, :, :]) if file_png is not None: if os.path.splitext(file_png)[1] != ".png": file_png += ".png" if verbose: print("Writing example png ({})...".format(file_png)) energy = np.sum(np.abs(kspace) ** 2, axis=(-1, -2, -4)) i_phase, i_echo, i_slice, i_z = np.where(energy == energy.max()) image_out = image[i_phase[0], i_echo[0], i_slice[0], :, i_z[0], :, :] image_out = np.sqrt(np.sum(np.abs(image_out) ** 2, axis=0)) image_out = image_out / np.max(image_out) * np.iinfo(np.uint8).max imageio.imwrite(file_png, image_out.astype(np.uint8)) if __name__ == "__main__": parser = argparse.ArgumentParser(description="Process ISMRMRD files") parser.add_argument("input", action="store", help="raw data in ISMRMRD format") parser.add_argument("-o", "--output", default="./", help="output directory (default: ./)") parser.add_argument("-p", "--png", default=None, help="png file name (default: None)") parser.add_argument("-s", "--suffix", default="", help="suffix to file output") parser.add_argument("-v", "--verbose", action="store_true", help="verbose printing (default: False)") args = parser.parse_args() dataset_to_cfl(args.output, args.input, suffix=args.suffix, file_png=args.png, verbose=args.verbose)
Python
2D
MRSRL/mridata-recon
fileio/__init__.py
.py
0
0
null
Python
2D
MRSRL/mridata-recon
fileio/cfl.py
.py
1,764
72
"""Read/write files for BART""" # This file is slightly modified from the file: # https://github.com/mrirecon/bart/blob/master/python/cfl.py # Copyright 2013-2015. The Regents of the University of California. # All rights reserved. Use of this source code is governed by # a BSD-style license which can be found in the LICENSE file. # # Authors: # 2013 Martin Uecker <uecker@eecs.berkeley.edu> # 2015 Jonathan Tamir <jtamir@eecs.berkeley.edu> import numpy as np def read_hdr(name, order='C'): """Read hdr file.""" # get dims from .hdr h = open(name + ".hdr", "r") h.readline() # skip line l = h.readline() h.close() dims = [int(i) for i in l.split()] if order == 'C': dims.reverse() return dims def read(name, order='C'): dims = read_hdr(name, order) # remove singleton dimensions from the end n = np.prod(dims) # dims_prod = np.cumprod(dims) # dims = dims[:np.searchsorted(dims_prod, n)+1] # load data and reshape into dims d = open(name + ".cfl", "r") a = np.fromfile(d, dtype=np.complex64, count=n) d.close() return a.reshape(dims, order=order) def readcfl(name): return read(name, order='F') def write(name, array, order='C'): h = open(name + ".hdr", "w") h.write('# Dimensions\n') if order=='C': for i in array.shape[::-1]: h.write("%d " % i) else: for i in (array.shape): h.write("%d " % i) h.write('\n') h.close() d = open(name + ".cfl", "w") if order=='C': array.astype(np.complex64).tofile(d) else: # tranpose for column-major order array.T.astype(np.complex64).tofile(d) d.close() def writecfl(name, array): write(name, array, order='F')
Python
2D
MRSRL/mridata-recon
mrirecon/fse.py
.py
2,670
73
"""General recon functions for FSE datasets.""" import numpy as np from mrirecon import fftc def _compute_coefficients_ahncho(kscalib): """Compute correction coefficients for FSE using Ahn-Cho method. kscalib: [..., channels, segments, kx] """ # offset start and end to avoid effects from fft wrap istart = 5 iend = -5 kscalib_shape = kscalib.shape num_segments = kscalib_shape[-2] num_kx = kscalib_shape[-1] kscalib = np.reshape(kscalib, [-1, num_segments, num_kx]) imcalib = fftc.ifftc(kscalib, axis=-1) imcalib_ref = np.conj(imcalib) * imcalib[:, :1, :] p1_calc = np.angle(np.mean(imcalib_ref[:, :, istart:iend] * np.conj(imcalib_ref[:, :, (istart+1):(iend+1)]), axis=-1)) p1_calc = np.expand_dims(p1_calc, axis=-1) x = np.arange(num_kx * 1.0) x = np.reshape(x, [1, 1, num_kx]) imcalib_cor1 = imcalib_ref * np.exp(1j * x * p1_calc) p0_calc = np.angle(np.mean(imcalib_cor1, axis=-1)) p1_calc = np.reshape(p1_calc, kscalib_shape[:-2] + p1_calc.shape[-2:]) p0_calc = np.reshape(p0_calc, kscalib_shape[:-2] + p0_calc.shape[-1:] + (1,)) return p0_calc, -p1_calc def phase_correction(kspace, echo_train, kspace_fse_calib, echo_train_fse_calib): """Perform linear phase correction for FSE scans kspace: [phases, echoes, slices, channels, kz, ky, kx] echo_train: [phases, echoes, slices, 1, segments, 1] """ x = np.arange(kspace.shape[-1] * 1.0) x = np.reshape(x, [1, 1, -1]) p0, p1 = _compute_coefficients_ahncho(kspace_fse_calib) ksx = fftc.ifftc(kspace, axis=-1) num_phases = ksx.shape[0] num_echoes = ksx.shape[1] num_slices = ksx.shape[2] num_kz = ksx.shape[-3] for i_phase in range(num_phases): for i_echo in range(num_echoes): for i_slice in range(num_slices): for i_kz in range(num_kz): ind = echo_train[i_phase, i_echo, i_slice, 0, i_kz, :, 0] ind_calib = echo_train_fse_calib[i_phase, i_echo, i_slice, 0, :, 0] ks_slice = ksx[i_phase, i_echo, i_slice, :, i_kz, :, :] p0_slice = p0[i_phase, i_echo, i_slice, :, ind_calib, :] p0_slice = np.transpose(p0_slice[ind, :, :], [1, 0, 2]) p1_slice = p1[i_phase, i_echo, i_slice, :, ind_calib, :] p1_slice = np.transpose(p1_slice[ind, :, :], [1, 0, 2]) ks_slice *= np.exp(1j * (p0_slice + x * p1_slice)) ksx[i_phase, i_echo, i_slice, :, i_kz, :, :] = ks_slice kspace_cor = fftc.fftc(ksx, axis=-1) return kspace_cor
Python
2D
MRSRL/mridata-recon
mrirecon/__init__.py
.py
0
0
null
Python
2D
MRSRL/mridata-recon
mrirecon/fftc.py
.py
1,755
77
try: import pyfftw.interfaces.numpy_fft as fft except: from numpy import fft import numpy as np def ifftnc(x, axes): tmp = fft.fftshift(x, axes=axes) tmp = fft.ifftn(tmp, axes=axes) return fft.ifftshift(tmp, axes=axes) def fftnc(x, axes): tmp = fft.fftshift(x, axes=axes) tmp = fft.fftn(tmp, axes=axes) return fft.ifftshift(tmp, axes=axes) def fftc(x, axis=0, do_orthonorm=True): if do_orthonorm: scale = np.sqrt(x.shape[axis]) else: scale = 1.0 return fftnc(x, (axis,)) / scale def ifftc(x, axis=0, do_orthonorm=True): if do_orthonorm: scale = np.sqrt(x.shape[axis]) else: scale = 1.0 return ifftnc(x, (axis,)) * scale def fft2c(x, order='C', do_orthonorm=True): if order == 'C': if do_orthonorm: scale = np.sqrt(np.prod(x.shape[-2:])) else: scale = 1.0 return fftnc(x, (-2, -1)) / scale else: if do_orthonorm: scale = np.sqrt(np.prod(x.shape[:2])) else: scale = 1.0 return fftnc(x, (0, 1)) / scale def ifft2c(x, order='C', do_orthonorm=True): if order == 'C': if do_orthonorm: scale = np.sqrt(np.prod(x.shape[-2:])) else: scale = 1.0 return ifftnc(x, (-2, -1)) * scale else: if do_orthonorm: scale = np.sqrt(np.prod(x.shape[:2])) else: scale = 1.0 return ifftnc(x, (0, 1)) * scale def fft3c(x, order='C'): if order == 'C': return fftnc(x, (-3, -2, -1)) else: return fftnc(x, (0, 1, 2)) def ifft3c(x, order='C'): if order == 'C': return ifftnc(x, (-3, -2, -1)) else: return ifftnc(x, (0, 1, 2))
Python
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
Library.m
.m
21,731
709
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Library load %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% DB = importdata('materialDB_ZB.csv',','); %DB = importdata('materialDB_WZ.csv',','); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%% Here is a patch to be able to load the tables in Matlab AND Octave %%%%%% % Matlab see the header in multiple cells while Octave see the header in one cell only %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='GaAs'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); GaAs = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 GaAs=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlAs'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlAs = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlAs=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='InAs'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); InAs = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 InAs=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='InSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); InSb = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 InSb=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='GaSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); GaSb = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 GaSb=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlSb = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlSb=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='InP'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); InP = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 InP=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='GaP'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); GaP = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 GaP=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlP'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlP = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlP=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='Si'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); Si = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 Si=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='Ge'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); Ge = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 Ge=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='SiGe'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); SiGe_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 SiGe_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlGaAs'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlGaAs_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlGaAs_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='InGaAs'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); InGaAs_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 InGaAs_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlInAs'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlInAs_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlInAs_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='GaInP'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); GaInP_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 GaInP_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlInP'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlInP_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlInP_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlGaP'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlGaP_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlGaP_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='GaInSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); GaInSb_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 GaInSb_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlInSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlInSb_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlInSb_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlGaSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlGaSb_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlGaSb_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='GaAsSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); GaAsSb_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 GaAsSb_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='InAsSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); InAsSb_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 InAsSb_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlAsSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlAsSb_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlAsSb_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='GaAsP'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); GaAsP_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 GaAsP_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='InAsP'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); InAsP_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 InAsP_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlAsP'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlAsP_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlAsP_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='GaPSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); GaPSb_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 GaPSb_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='InPSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); InPSb_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 InPSb_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% M='AlPSb'; if length(DB.textdata(1,:))==1 %% Octave data load DB.textdata{1,1}=[DB.textdata{1,1} ',']; % patch, add a comma "," at the end idxM=strfind(DB.textdata{1,1},','); idx=strfind(DB.textdata{1,1},[',' M ',']); idxM=find(idxM==idx); AlPSb_bowing = DB.data(:,idxM)'; else %% Matlab data load for i=1:length(DB.textdata(1,:)) idx=strcmp(DB.textdata{1,i},M); if idx==1 AlPSb_bowing=DB.data(:,i-1)'; % break % removing the break makes it slower but more compatible between Matlab and Octave end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
computesISBdipoles.m
.m
1,345
30
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%% Computes ISB dipoles %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Take care! Some people use meff inside the oscillator strenght f % Actually, meff has sens in an infinite QW because there is a single mass value % but NOT in multi-QW structure with various materials % https://www.nextnano.com/nextnano3/tutorial/1Dtutorial_IntrabandTransitions.htm % https://www.nextnano.com/nextnano3/tutorial/1Dtutorial_InGaAs_MQWs.htm %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% z_dipole_c = zeros(length(Ec),length(Ec)); f_dipole_c = zeros(length(Ec),length(Ec)); for i=1:length(Ec) for j=1:length(Ec) if j>i z_dipole_c(i,j) = abs( trapz( z , psic(:,i).*z'.*psic(:,j) ) ); f_dipole_c(i,j) = 2*m0/hbar^2 * ( Ec(j)-Ec(i) )* e * z_dipole_c(i,j)^2 ; end end end z_dipole_c = z_dipole_c + z_dipole_c.'; f_dipole_c = f_dipole_c + f_dipole_c.'; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
SchrodingerPoisson1D_CB_Kane_Main.m
.m
18,760
492
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%% last update 20December2021, lne %%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % This program solves the Schrodinger-Poisson equation in the conduction band % for any heterostructures in the Zinc-Blend material. % The program solves the Schrodinger equation with m(E,z) using different algorithms: % -> FDM (Finite Difference Method) % and / or % -> Scaning/Shooting method % The non-parabolicity is implemented via the kp model using the Kane model % A strain model is included. It basically shifts the conduction band edge % The strain is mainly interesting for InGaAs/GaAs heterostructures % The non-parabolicity is also included into the density of states for the Poisson solver. % It follows the book of Paul Harrison and actually makes the 2d density of states not constant % -> Additionnal material can be added in the "materialDB_ZB.csv" file % -> II-VI and cubic nitride material parameters are available but should % be grabt in the "Library.m" file %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % If the code doesn t converge: % -> decrease the doping % -> increase the resolution dz % -> increase the temperature (T=0K is very bad while T=10K is already much better) % -> increase the amount of loops, Nloops %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% clear all close all clc %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Constants %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% h = 6.62606896E-34; %% Planck constant [J.s] hbar = h/(2*pi); e = 1.602176487E-19; %% electron charge [C] m0 = 9.10938188E-31; %% electron mass [kg] Epsi0= 8.854187817620E-12; %% Vaccum dielectric constant [F/m] kB = 1.3806488E-23; %% Boltzmann's constant [J/K] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Nloops = 5; % number of loops model = 2; % 1=shooting; 2=Kane StrainModel = 0; % Activate Strain model n = 3; % number of solution asked per model ScF = 0.1; % scaling factor to plot the wave function [Without Dimension] dz = 1e-10; % resolution of the grid [m] T = 300; % Temperature [Kelvin] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% DisplayResults = 1; % Switch to print or not the ISB dipoles on the shell plot_density = 1; % Activate the plot 0 or 1 plot_convergence = 0; % Activate the plot 0 or 1 plot_field = 0; % Activate the plot 0 or 1 plot_Vbending = 0; % Activate the plot 0 or 1 plot_mass = 0; % Activate the plot 0 or 1 plot_ro = 0; % Activate the plot 0 or 1 plot_Epsi = 0; % Activate the plot 0 or 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Library; % load material parameter DB from "materialDB_ZB.csv" ExtractParameters; % extract parameter from the Library TernaryAlloy; % compute the ternary alloy %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%% import the layer structure file %%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % first column is the material used from the "library" % second column is the length of the layer in nm % third column is the n doping volumique of that layer in 1e18 cm-3 % You have to put a resonable amount of doping! Otherwise, it will diverge % and you will have to damp it more by increasing the number of loops ! %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% input_file; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%% NOTHING TO CHANGE ANYMORE !!! %%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%% Grabbing the parameters %%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% zt = M(:,end-1)*1e-9; % conversion of the length from nm to meter Dopt = M(:,end)*1e18*1e6; % n doping conversion from cm-3 to m-3 Egt = M(:,idx_Eg6c) - (M(:,idx_alphaG)*T^2) ./ (T+M(:,idx_betaG)); %Eg = Eg0 - (a*T.^2)./(T + b); CBOt = Egt+M(:,idx_VBO); % CBO form band gap difference and temperature Dsot = M(:,idx_Dso); % Spin-Orbit shift band parameter EPt_K= M(:,idx_EP_K); % EP Kane Epsit= M(:,idx_Epsi); % Epsilon; dielectric constant %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Strain Model %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% at = M(:,idx_a); % lattice parameter act = M(:,idx_ac); % Conduction band strain offset parameter avt = M(:,idx_av); % Valence band strain offset parameter bvt = M(:,idx_bv); % Valence band strain offset parameter c11t = M(:,idx_c11); % strain parameter c12t = M(:,idx_c12); % strain parameter a0 = substrate(idx_a); if StrainModel == 1 exxt = (a0-at)/a0; % eyyt = exxt; ezzt = -2*c12t./c11t.*exxt; else exxt = (a0-at)/a0 * 0; % eyyt = exxt; ezzt = -2*c12t./c11t.*exxt; end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Discretisation %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % here, I descretize the grid z, the potential V0 and that values that are needed z=0; V0=CBOt(1); Eg=Egt(1); Dso=Dsot(1); EP_K=EPt_K(1); Dop=Dopt(1); Epsi=Epsit(1); ac=act(1); av=avt(1); bv=bvt(1); exx=exxt(1); ezz=ezzt(1); for i=1:length(zt) t=zt(i); zv= (z(end)+dz): dz : (z(end)+dz)+t; z=[z zv]; V0 = [ V0 ones(size(zv)) * CBOt(i) ]; Eg = [ Eg ones(size(zv)) * Egt(i) ]; Dop = [ Dop ones(size(zv)) * Dopt(i) ]; Epsi= [ Epsi ones(size(zv)) * Epsit(i) ]; EP_K= [ EP_K ones(size(zv)) * EPt_K(i) ]; Dso = [ Dso ones(size(zv)) * Dsot(i) ]; ac = [ ac ones(size(zv)) * act(i) ]; av = [ av ones(size(zv)) * avt(i) ]; bv = [ bv ones(size(zv)) * bvt(i) ]; exx = [ exx ones(size(zv)) * exxt(i) ]; ezz = [ ezz ones(size(zv)) * ezzt(i) ]; end V0=V0-min(V0); % Shift the band in order to get the bottom of the well at zero V0=(F0*z)+V0; % adding the electric field to the potential Ltot=z(end)-z(1); eyy = exx; DCBO = -abs(ac).*(exx+eyy+ezz) ; % shift of the CB due to strain Ntott= Dopt.*zt; Ntot = sum(Ntott); % total number of charges %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% dE1=1e-3; dE2=1e-2; if StrainModel == 0 E1 = min(V0):dE1:min(V0)+0.2 ; E2 = E1(end):dE2:max(V0); En = sort([E1 E2]); else E1 = min(V0+DCBO):dE1:min(V0+DCBO)+0.2 ; E2 = E1(end):dE2:max(V0+DCBO); En = sort([E1 E2]); end if model==1 EEn = repmat(En', [1 length(z)]); V0mat = repmat(V0 , [length(En) 1]); Egmat = repmat(Eg , [length(En) 1]); Dsomat = repmat(Dso, [length(En) 1]); EPmat = repmat(EP_K ,[length(En) 1]); meK2 = 1 ./ ( 1 + EPmat ./ (EEn + Egmat-V0mat+1/3*Dsomat) ); % m(z,E) in the 2 bands Kane s model % here, I linearize the mass versus the energy meff0 = 1 ./ ( 1 + EPmat ./ (Egmat-V0mat+1/3*Dsomat) ); alpha = ( meK2(end,:)./meff0(end,:) - 1 ) ./ EEn(end,:); alphamat= repmat(alpha , [length(En) 1] ); meK2lin = meff0.*(1+alphamat.*EEn); % => linearized mass end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%% Starting of the Poisson s loop %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Vs=zeros(size(z)); Vsold=Vs; ntot=0; nloop=1; ErrVec=1; sumVtotVec=1; while nloop<Nloops nloop x = 0.5; Vbending=Vs*x + Vsold*(1-x); Vtot=V0+Vbending; %%%%%%%%%%%%%%%%%%%%%%%%%%%% schrodinger solver %%%%%%%%%%%%%%%%%%%%%%%%%%%% if model==1 dE=0.002; precision=1e-7; Vtotmat = repmat(Vtot, [length(En) 1]); meK2 = 1 ./ ( 1 + EPmat ./ (EEn + Egmat-Vtotmat+1/3*Dsomat) ); % m(z,E) in the 2 bands Kane s model [Ec,psic] = Schrod_Nbands_shoot_f(z,Vtot+DCBO,meK2,n,En,dE,precision); elseif model==2 [Ec,psic] = Schrod_2bands_Kane_f(z,Vtot,Eg,EP_K,Dso,n,ac,av,bv,exx,ezz); end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Here, I re-define the energy grid in order optimize the meshing dE1=1e-4; dE2=1e-2; if StrainModel == 0 E1 = Ec(1):dE1:Ec(1)+0.1 ; E2 = E1(end):dE2:max(Vtot); En=sort([E1 E2]); else E1 = Ec(1):dE1:Ec(1)+0.1 ; E2 = E1(end):dE2:max(Vtot+DCBO); En=sort([E1 E2]); end EEn = repmat(En' , [1 length(z) ]); V0mat = repmat(V0 , [length(En) 1]); Vtotmat = repmat(Vtot, [length(En) 1]); Egmat = repmat(Eg , [length(En) 1]); Dsomat = repmat(Dso , [length(En) 1]); EPmat = repmat(EP_K, [length(En) 1]); meK2 = 1 ./ ( 1 + EPmat ./ (EEn + Egmat-Vtotmat+1/3*Dsomat) ); % m(z,E) in the 2 bands Kane s model % here, I linearize the mass versus the energy meff0 = 1 ./ ( 1 + EPmat ./ (Egmat-Vtotmat+1/3*Dsomat) ); alpha = ( meK2(end,:)./meff0(end,:) - 1 ) ./ EEn(end,:); alphamat= repmat(alpha , [length(En) 1] ); meK2lin = meff0.*(1+alphamat.*EEn); % => linearized mass %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Paul Harrisson % Quantum Wells, Wires and Dots. % 4th edition (2016), % chap 2 : "Solutions to Schrodinger's equation" % 2.42: "Two-dimensional systems" page 31 meffro = meff0.*(1+2*alphamat.*EEn); % => linearized mass ro=[]; for i=1:length(Ec) %ro( En>Ec(i),:,i) = e* meK2(En>Ec(i),:) * m0/(pi*hbar^2); %ro( En>Ec(i),:,i) = e*meK2lin(En>Ec(i),:) * m0/(pi*hbar^2); ro( En>Ec(i),:,i) = e*meffro(En>Ec(i),:) * m0/(pi*hbar^2); ro( En<Ec(i),:,i) = 0; end % here is what make the code very slow if Ntot==0 idx=find(min(CBOt)==CBOt); Ef=-Egt(idx(1))/2; NN=0*Ec'; roEf=ro*0; else [Ef,NN,roEf]=find_Ef_f(z,Ec,psic,En,ro,Ntot,T); end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% ntot2 = repmat(NN,[length(z) 1]).*abs(psic).^2 ; ntot = sum(ntot2,2)' - Dop; % remove the charge positives (ions) %%%%%%%%%%%%%%%%%%%%%%%%%% Electrical Field %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% F = e*cumtrapz(z,ntot)./(Epsi0*Epsi); MF = trapz(z,F)/(z(end)-z(1)); % MF=mean(F) on a nonlinear grid z F = F-MF; %%%%%%%%%%%%%%%%%%%%%%%%%%% New Potential %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Vsold = Vs; Vs = -cumsum(F)*dz; %%%%%%%%%%%%%%%%%%%%%%%%%%%% Convergence %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Err = abs( 1 - sumVtotVec(end)/sum(Vs) ); sumVtotVec(nloop) = sum(Vs); ErrVec = [ErrVec Err]; nloop=nloop+1; end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Display Results %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% scale_PSI; computesISBdipoles; if DisplayResults == 1 PrintResults; end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% figures %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %figure('position',[-3500 100 1000 700],'color','w'); figure('position',[10 100 1000 700],'color','w'); subplot(1,1,1,'fontsize',15) hold on; grid on; box on; col=colormap(jet); if plot_density==1 grid off pcolor(z*1e9,En,sum(ROEf,3)*1e-6 ) set(gca,'color',col(1,:)) shading flat hcb=colorbar; caxis([0 max(max(sum(ROEf,3)*1e-6))+1]) title(hcb,'\fontsize{8}cm-3') if StrainModel == 0 plot(z*1e9,V0, 'w--','linewidth',1) plot(z*1e9,Vtot,'w-' ,'linewidth',1) else plot(z*1e9,V0+DCBO, 'k--','linewidth',1) plot(z*1e9,Vtot+DCBO,'k-' ,'linewidth',1) end elseif plot_density==0 if StrainModel == 0 plot(z*1e9,V0, 'b--','linewidth',1) plot(z*1e9,Vtot,'b-' ,'linewidth',1) else plot(z*1e9,V0+DCBO, 'k--','linewidth',1) plot(z*1e9,Vtot+DCBO,'k-' ,'linewidth',1) end end for i=1:length(Ec) plot(z*1e9,PSIc(:,i),'color','r','linewidth',1) end plot([z(1) z(end)]*1e9,[1 1]*Ef,'g','linewidth',1) text(z(end)*1e9*0.95,Ef+0.01,'\color{green}Fermi') xlabel('z (nm)') ylabel('Energy (eV)') xlim([0 z(end)*1e9]) if StrainModel == 1 title(strcat('T=',num2str(T),'K ; dz=',num2str(dz*1e9),'nm; Ntot=',num2str(Ntot*1e-4,'%.1e'),'cm-2; with STRAIN')) else title(strcat('T=',num2str(T),'K ; dz=',num2str(dz*1e9),'nm; Ntot=',num2str(Ntot*1e-4,'%.1e'),'cm-2; without STRAIN')) end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_convergence==1 figure('color','w') semilogy(1:nloop,ErrVec,'bo-') hold on; grid on; box on; xlabel('Cycles') ylabel('Convergence (norm. units)') end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_field==1 figure('color','w') hold on; grid on; box on; [AX,H1,H2]=plotyy(z*1e9,F*1e-2*1e-3,z*1e9,Dop*1e-18*1e-6); set(H1,'color','r') set(H2,'color','b') xlabel('z (nm)') ylabel(AX(1),'E- field (kV/cm)','color','red') ylabel(AX(2),'Doping (1e18 cm-3)','color','blue') set(AX(1),'ycolor','red') set(AX(2),'ycolor','blue') end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_Vbending==1 figure('color','w') hold on; grid on; box on; [AX,H1,H2]=plotyy(z*1e9,Vbending,z*1e9,ntot*1e-18*1e-6); set(H1,'color','r') set(H2,'color','b') xlabel('z (nm)') ylabel(AX(1),'Vbending (eV)','color','red') ylabel(AX(2),'ntot (1e18 cm-3)','color','blue') set(AX(1),'ycolor','red') set(AX(2),'ycolor','blue') end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_mass==1 figure('color','w') subplot(1,2,1) hold on; grid on; box on; pcolor(z*1e9,En,meK2) colormap(jet) hcb=colorbar; title(hcb,'\fontsize{8}meff') shading flat xlabel('z (nm)') ylabel('Energy (eV)') title('meK2') xlim([0 z(end)*1e9]) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% subplot(1,2,2) hold on; grid on; box on; pcolor(z*1e9,En,meK2lin) colormap(jet) hcb=colorbar; title(hcb,'\fontsize{8}meff') shading flat xlabel('z (nm)') ylabel('Energy (eV)') title('meK2-lin') xlim([0 z(end)*1e9]) % figure % hold on;grid on; % idx=250; % plot(En,meK2(:,idx),'b') % plot(En,meK2lin(:,idx),'r') end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_ro==1 figure('color','w') hold on; grid on; box on; surf(z*1e9,En,sum(ro,3)) view(45,30) colormap(jet) hcb=colorbar; title(hcb,'\fontsize{8}ro') shading flat xlabel('z (nm)') ylabel('Energy (eV)') xlim([0 z(end)*1e9]) end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_Epsi==1 figure('color','w') hold on; grid on; box on; plot(z*1e9,Epsi,'.-') xlabel('z (nm)') ylabel('Epsilon') xlim([0 z(end)*1e9]) end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
scale_PSI.m
.m
815
17
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%% Scaling and shifting the wavefunctions %%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% for i=1:length(Ec) PSIc(:,i)=abs(psic(:,i)).^2/max(abs(psic(:,i)).^2)*ScF + Ec(i); % normalisation for the plotting end % ROEf is only build to plot the density of electron in a 2D map for ii=1:length(Ec) psic_M = repmat(psic(:,ii),[1,length(En)])'; ROEf(:,:,ii) = roEf(:,:,ii) .* abs(psic_M.^2); end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
SchrodingerPoisson1D_CB_Main.m
.m
17,453
457
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%% last update 25December2021, lne %%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % This program solves the Schrodinger-Poisson equation in the conduction band % for any heterostructures. % The program solves the Schrodinger equation with m(E,z) using the Scaning/Shooting % method. The non-parabolicity is implemented via the alpha parameter, alpha=1/Egap; % meff(E)=meff(0)*(1+alpha*E); % It can be easily modified by tuning the alpha in each layer => line 109 % A strain model is included. It basically shifts the conduction band edge % The strain is mainly interesting for InGaAs/GaAs heterostructures % The non-parabolicity is also included into the density of states for the Poisson solver. % It follows the book of Paul Harrison and actually makes the 2d density of states not constant % -> Additionnal material can be added in the "materialDB_ZB.csv" file % -> II-VI and cubic nitride material parameters are available but should % be grabt in the "Library.m" file %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % If the code doesn t converge: % -> decrease the doping % -> increase the resolution dz % -> increase the temperature (T=0K is very bad while T=10K is already much better) % -> increase the amount of loops, Nloops %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% clear all close all clc %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Constants %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% h = 6.62606896E-34; %% Planck constant [J.s] hbar = h/(2*pi); e = 1.602176487E-19; %% electron charge [C] m0 = 9.10938188E-31; %% electron mass [kg] Epsi0= 8.854187817620E-12; %% Vaccum dielectric constant [F/m] kB = 1.3806488E-23; %% Boltzmann's constant [J/K] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Nloops = 5; % number of loops StrainModel = 0; % Activate Strain model n = 4; % number of solution asked per model ScF = 0.1; % scaling factor to plot the wave function [Without Dimension] dz = 1e-10; % resolution of the grid [m] T = 300; % Temperature [Kelvin] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% DisplayResults = 1; % Switch to print or not the ISB dipoles on the shell plot_density = 1; % Activate the plot 0 or 1 plot_convergence = 0; % Activate the plot 0 or 1 plot_field = 0; % Activate the plot 0 or 1 plot_Vbending = 0; % Activate the plot 0 or 1 plot_mass = 0; % Activate the plot 0 or 1 plot_ro = 0; % Activate the plot 0 or 1 plot_Epsi = 0; % Activate the plot 0 or 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Library; % load material parameter DB from "materialDB_ZB.csv" ExtractParameters; % extract parameter from the Library TernaryAlloy; % compute the ternary alloy %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%% import the layer structure file %%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % first column is the material used from the "library" % second column is the length of the layer in nm % third column is the n doping volumique of that layer in 1e18 cm-3 % You have to put a resonable amount of doping! Otherwise, it will diverge % and you will have to damp it more by increasing the number of loops ! %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% input_file; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%% NOTHING TO CHANGE ANYMORE !!! %%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%% Grabbing the parameters %%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% zt = M(:,end-1)*1e-9; % conversion of the length from nm to meter Dopt = M(:,end)*1e18*1e6; % n doping conversion from cm-3 to m-3 Egt = M(:,idx_Eg6c) - (M(:,idx_alphaG)*T^2) ./ (T+M(:,idx_betaG)); %Eg = Eg0 - (a*T.^2)./(T + b); CBOt = Egt+M(:,idx_VBO); % CBO form band gap difference and temperature %Dsot = M(:,idx_Dso); % Spin-Orbit shift band parameter %EPt_K= M(:,idx_EP_K); % EP Kane Epsit = M(:,idx_Epsi); % Epsilon; dielectric constant mefft = M(:,idx_me); % electron effective mass at band edge alphat= 1./Egt; % non-parabolicity parameter => can be adjusted! %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Strain Model %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% at = M(:,idx_a); % lattice parameter act = M(:,idx_ac); % Conduction band strain offset parameter avt = M(:,idx_av); % Valence band strain offset parameter bvt = M(:,idx_bv); % Valence band strain offset parameter c11t = M(:,idx_c11); % strain parameter c12t = M(:,idx_c12); % strain parameter a0 = substrate(idx_a); if StrainModel == 1 exxt = (a0-at)/a0; % eyyt = exxt; ezzt = -2*c12t./c11t.*exxt; else exxt = (a0-at)/a0 * 0; % eyyt = exxt; ezzt = -2*c12t./c11t.*exxt; end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Discretisation %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % here, I descretize the grid z, the potential V0 and that values that are needed z=0; V0=CBOt(1); Eg=Egt(1); meff=mefft(1); alpha=alphat(1); Dop=Dopt(1); Epsi=Epsit(1); ac=act(1); av=avt(1); bv=bvt(1); exx=exxt(1); ezz=ezzt(1); for i=1:length(zt) t = zt(i); zv = (z(end)+dz): dz : (z(end)+dz)+t; z = [z zv]; V0 = [ V0 ones(size(zv)) * CBOt(i) ]; Eg = [ Eg ones(size(zv)) * Egt(i) ]; Dop = [ Dop ones(size(zv)) * Dopt(i) ]; Epsi = [ Epsi ones(size(zv)) * Epsit(i) ]; meff = [ meff ones(size(zv)) * mefft(i) ]; alpha= [ alpha ones(size(zv)) * alphat(i)]; ac = [ ac ones(size(zv)) * act(i) ]; av = [ av ones(size(zv)) * avt(i) ]; bv = [ bv ones(size(zv)) * bvt(i) ]; exx = [ exx ones(size(zv)) * exxt(i) ]; ezz = [ ezz ones(size(zv)) * ezzt(i) ]; end V0=V0-min(V0); % Shift the band in order to get the bottom of the well at zero V0=(F0*z)+V0; % adding the electric field to the potential Ltot=z(end)-z(1); eyy = exx; DCBO = -abs(ac).*(exx+eyy+ezz) ; % shift of the CB due to strain Ntott=Dopt.*zt; Ntot=sum(Ntott); % total number of charges %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% dE1=1e-3; dE2=1e-2; if StrainModel == 0 E1 = min(V0):dE1:min(V0)+0.2 ; E2 = E1(end):dE2:max(V0); En=sort([E1 E2]); else E1 = min(V0+DCBO):dE1:min(V0+DCBO)+0.2 ; E2 = E1(end):dE2:max(V0+DCBO); En=sort([E1 E2]); end EEn = repmat(En', [1 length(z)]); V0mat = repmat(V0 , [length(En) 1]); Egmat = repmat(Eg , [length(En) 1]); meffmat = repmat(meff , [length(En) 1]); alphamat= repmat(alpha, [length(En) 1]); melin = meffmat.*(1+alphamat.*(EEn-V0mat)); % => linearized mass %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%% Starting of the Poisson s loop %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Vs=zeros(size(z)); Vsold=Vs; ntot=0; nloop=1; ErrVec=1; sumVtotVec=1; while nloop<Nloops nloop x = 0.5; Vbending=Vs*x + Vsold*(1-x); Vtot=V0+Vbending; %%%%%%%%%%%%%%%%%%%%%%%%%%%% schrodinger solver %%%%%%%%%%%%%%%%%%%%%%%%%%%% dE=0.002; precision=1e-7; Vtotmat = repmat(Vtot, [length(En) 1]); melin = meffmat.*(1+alphamat.*(EEn-Vtotmat)); % => linearized mass [Ec,psic] = Schrod_Nbands_shoot_f(z,Vtot+DCBO,melin,n,En,dE,precision); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Here, I re-define the energy grid in order optimize the meshing dE1=1e-4; dE2=1e-2; if StrainModel == 0 E1 = Ec(1):dE1:Ec(1)+0.1 ; E2 = E1(end):dE2:max(Vtot); En=sort([E1 E2]); else E1 = Ec(1):dE1:Ec(1)+0.1 ; E2 = E1(end):dE2:max(Vtot+DCBO); En=sort([E1 E2]); end EEn = repmat(En' , [1 length(z) ]); V0mat = repmat(V0 , [length(En) 1]); Vtotmat = repmat(Vtot, [length(En) 1]); Egmat = repmat(Eg , [length(En) 1]); meffmat = repmat(meff ,[length(En) 1]); alphamat= repmat(alpha, [length(En) 1]); melin = meffmat.*(1+alphamat.*(EEn-Vtotmat)); % => linearized mass %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Paul Harrisson % Quantum Wells, Wires and Dots. % 4th edition (2016), % chap 2 : "Solutions to Schrodinger's equation" % 2.42: "Two-dimensional systems" page 31 meffro = meffmat.*(1+2*alphamat.*EEn); % => linearized mass ro=[]; for i=1:length(Ec) ro( En>Ec(i),:,i) = e*meffro(En>Ec(i),:) * m0/(pi*hbar^2); ro( En<Ec(i),:,i) = 0; end % here is what make the code very slow if Ntot==0 idx=find(min(CBOt)==CBOt); Ef=-Egt(idx(1))/2; NN=0*Ec'; roEf=ro*0; else [Ef,NN,roEf]=find_Ef_f(z,Ec,psic,En,ro,Ntot,T); end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% ntot2 = repmat(NN,[length(z) 1]).*abs(psic).^2 ; ntot = sum(ntot2,2)' - Dop; % remove the charge positives (ions) %%%%%%%%%%%%%%%%%%%%%%%%%% Electrical Field %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% F = e*cumtrapz(z,ntot)./(Epsi0*Epsi); MF = trapz(z,F)/(z(end)-z(1)); % MF=mean(F) on a nonlinear grid z F = F-MF; %%%%%%%%%%%%%%%%%%%%%%%%%%% New Potential %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Vsold = Vs; Vs = -cumsum(F)*dz; %%%%%%%%%%%%%%%%%%%%%%%%%%%% Convergence %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Err = abs( 1 - sumVtotVec(end)/sum(Vs) ); sumVtotVec(nloop) = sum(Vs); ErrVec = [ErrVec Err]; nloop=nloop+1; end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Display Results %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% scale_PSI; computesISBdipoles; if DisplayResults == 1 PrintResults; end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% figures %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %figure('position',[-3500 100 1000 700],'color','w'); figure('position',[10 100 1000 700],'color','w'); subplot(1,1,1,'fontsize',15) hold on; grid on; box on; col=colormap(jet); if plot_density==1 grid off pcolor(z*1e9,En,sum(ROEf,3)*1e-6 ) set(gca,'color',col(1,:)) shading flat hcb=colorbar; caxis([0 max(max(sum(ROEf,3)*1e-6))+1]) title(hcb,'\fontsize{8}cm-3') if StrainModel == 0 plot(z*1e9,V0, 'w--','linewidth',1) plot(z*1e9,Vtot,'w-' ,'linewidth',1) else plot(z*1e9,V0+DCBO, 'k--','linewidth',1) plot(z*1e9,Vtot+DCBO,'k-' ,'linewidth',1) end elseif plot_density==0 if StrainModel == 0 plot(z*1e9,V0, 'b--','linewidth',1) plot(z*1e9,Vtot,'b-' ,'linewidth',1) else plot(z*1e9,V0+DCBO, 'k--','linewidth',1) plot(z*1e9,Vtot+DCBO,'k-' ,'linewidth',1) end end for i=1:length(Ec) plot(z*1e9,PSIc(:,i),'color','r','linewidth',1) end plot([z(1) z(end)]*1e9,[1 1]*Ef,'g','linewidth',1) text(z(end)*1e9*0.95,Ef+0.01,'\color{green}Fermi') xlabel('z (nm)') ylabel('Energy (eV)') xlim([0 z(end)*1e9]) if StrainModel == 1 title(strcat('T=',num2str(T),'K ; dz=',num2str(dz*1e9),'nm; Ntot=',num2str(Ntot*1e-4,'%.1e'),'cm-2; with STRAIN')) else title(strcat('T=',num2str(T),'K ; dz=',num2str(dz*1e9),'nm; Ntot=',num2str(Ntot*1e-4,'%.1e'),'cm-2; without STRAIN')) end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_convergence==1 figure('color','w') semilogy(1:nloop,ErrVec,'bo-') hold on; grid on; box on; xlabel('Cycles') ylabel('Convergence (norm. units)') end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_field==1 figure('color','w') hold on; grid on; box on; [AX,H1,H2]=plotyy(z*1e9,F*1e-2*1e-3,z*1e9,Dop*1e-18*1e-6); set(H1,'color','r') set(H2,'color','b') xlabel('z (nm)') ylabel(AX(1),'E- field (kV/cm)','color','red') ylabel(AX(2),'Doping (1e18 cm-3)','color','blue') set(AX(1),'ycolor','red') set(AX(2),'ycolor','blue') end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_Vbending==1 figure('color','w') hold on; grid on; box on; [AX,H1,H2]=plotyy(z*1e9,Vbending,z*1e9,ntot*1e-18*1e-6); set(H1,'color','r') set(H2,'color','b') xlabel('z (nm)') ylabel(AX(1),'Vbending (eV)','color','red') ylabel(AX(2),'ntot (1e18 cm-3)','color','blue') set(AX(1),'ycolor','red') set(AX(2),'ycolor','blue') end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_mass==1 figure('color','w') subplot(1,1,1) hold on; grid on; box on; pcolor(z*1e9,En,melin) colormap(jet) hcb=colorbar; title(hcb,'\fontsize{8}meff') shading flat xlabel('z (nm)') ylabel('Energy (eV)') title('me-lin') xlim([0 z(end)*1e9]) % figure % hold on;grid on; % idx=250; % plot(En,melin(:,idx),'r') end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_ro==1 figure('color','w') hold on; grid on; box on; surf(z*1e9,En,sum(ro,3)) view(45,30) colormap(jet) hcb=colorbar; title(hcb,'\fontsize{8}ro') shading flat xlabel('z (nm)') ylabel('Energy (eV)') xlim([0 z(end)*1e9]) end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if plot_Epsi==1 figure('color','w') hold on; grid on; box on; plot(z*1e9,Epsi,'.-') xlabel('z (nm)') ylabel('Epsilon') xlim([0 z(end)*1e9]) end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
TernaryAlloy.m
.m
5,184
123
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%% III-V Ternary alloys on InP %%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% x=0.53; InGaAs = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; % In0.53Ga0.47As is lattice matched on InP x=0.37; InGaAs37 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; % In0.37Ga0.63As STRAINED on InP x=0.45; InGaAs45 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; % In0.45Ga0.55As STRAINED on InP x=0.715; InGaAs71 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; % In0.715Ga0.285As STRAINED on InP %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% x=0.48; AlInAs = x*AlAs + (1-x)*InAs - x*(1-x)*AlInAs_bowing; % Al0.48In0.52As is lattice matched on InP x=0.80; AlInAs80 = x*AlAs + (1-x)*InAs - x*(1-x)*AlInAs_bowing; % Al0.80In0.20As STRAINED on InP %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% x=0.51; GaAsSb = x*GaAs + (1-x)*GaSb - x*(1-x)*GaAsSb_bowing; % GaAs0.51Sb0.49 is lattice matched on InP % QCL: Al-fee by G. Strasser's group in Vienna, JVSTB, 32. 02C104 (2014) % I did adjust VBO_bowing in order to get CBO=0.36eV between InGaAs/GaAsSb x=0.56; AlAsSb = x*AlAs + (1-x)*AlSb - x*(1-x)*AlAsSb_bowing; % AlAs0.56Sb0.44 is lattice matched on InP % the X-valley is 0.5eV above Gamma point (indirect bandgap material) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%% III-V Ternary alloys on GaAs %%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% x=0.10; AlGaAs_bowing(idx_Eg6c) = -0.127 + 1.310*x; % from Vurgaftman AlGaAs_bowing(idx_VBO) = 0.27 - 1.05*x; % to adjust VBO in order to fit CBO from THz QCL AlGaAs10 = x*AlAs + (1-x)*GaAs - x*(1-x)*AlGaAs_bowing; x=0.15; AlGaAs_bowing(idx_Eg6c) = -0.127 + 1.310*x; % from Vurgaftman AlGaAs_bowing(idx_VBO) = 0.27 - 1.05*x; % to adjust VBO in order to fit CBO from THz QCL AlGaAs15 = x*AlAs + (1-x)*GaAs - x*(1-x)*AlGaAs_bowing; x=0.20; AlGaAs_bowing(idx_Eg6c) = -0.127 + 1.310*x; % from Vurgaftman AlGaAs_bowing(idx_VBO) = 0.27 - 1.05*x; % to adjust VBO in order to fit CBO from THz QCL AlGaAs20 = x*AlAs + (1-x)*GaAs - x*(1-x)*AlGaAs_bowing; x=0.30; AlGaAs_bowing(idx_Eg6c) = -0.127 + 1.310*x; % from Vurgaftman AlGaAs_bowing(idx_VBO) = 0.27 - 1.05*x; % to adjust VBO in order to fit CBO from THz QCL AlGaAs30 = x*AlAs + (1-x)*GaAs - x*(1-x)*AlGaAs_bowing; x=0.40; AlGaAs_bowing(idx_Eg6c) = -0.127 + 1.310*x; % from Vurgaftman AlGaAs_bowing(idx_VBO) = 0.27 - 1.05*x; % to adjust VBO in order to fit CBO from THz QCL AlGaAs40 = x*AlAs + (1-x)*GaAs - x*(1-x)*AlGaAs_bowing; x=0.50; AlGaAs_bowing(idx_Eg6c) = -0.127 + 1.310*x; % from Vurgaftman AlGaAs_bowing(idx_VBO) = 0.27 - 1.05*x; % to adjust VBO in order to fit CBO from THz QCL AlGaAs50 = x*AlAs + (1-x)*GaAs - x*(1-x)*AlGaAs_bowing; x=0.05; InGaAs05 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; x=0.06; InGaAs06 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; x=0.07; InGaAs07 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; x=0.08; InGaAs08 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; x=0.09; InGaAs09 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; x=0.10; InGaAs10 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; x=0.15; InGaAs15 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; x=0.20; InGaAs20 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; x=0.25; InGaAs25 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; x=0.30; InGaAs30 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; x=0.40; InGaAs40 = x*InAs + (1-x)*GaAs - x*(1-x)*InGaAs_bowing; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%% III-V Ternary alloys on InAs/GaSb/AlSb %%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% x=0.25; GaInSb25 = x*InSb + (1-x)*GaSb - x*(1-x)*GaInSb_bowing; x=0.35; GaInSb35 = x*InSb + (1-x)*GaSb - x*(1-x)*GaInSb_bowing; x=0.09; GaAsSb91 = x*GaAs + (1-x)*GaSb - x*(1-x)*GaAsSb_bowing; % GaAs0.09Sb0.91 is lattice matched on InAs x=0.91; InAsSb = x*InAs + (1-x)*InSb - x.*(1-x)*InAsSb_bowing; % In0.09As0.91Sb is lattice matched on GaSb, Eg~0.3eV x=0.5;y=0.08; AlGaAsSb = x*GaSb + (1-x)*(y*AlAs + (1-y)*AlSb - y*(1-y)*AlAsSb_bowing); % high barrier lattice matched on GaSb %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%% IV-IV Ternary alloys %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% x=0.25; SiGe25 = x*Ge + (1-x)*Si - x*(1-x)*SiGe_bowing; x=0.3; SiGe30 = x*Ge + (1-x)*Si - x*(1-x)*SiGe_bowing; x=0.35; SiGe35 = x*Ge + (1-x)*Si - x*(1-x)*SiGe_bowing; x=0.4; SiGe40 = x*Ge + (1-x)*Si - x*(1-x)*SiGe_bowing; x=0.5; SiGe50 = x*Ge + (1-x)*Si - x*(1-x)*SiGe_bowing; x=0.8; SiGe80 = x*Ge + (1-x)*Si - x*(1-x)*SiGe_bowing;
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
input_file.m
.m
1,620
46
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%% Layers Structure %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % first column is the material used from the "library" % second column is the length of the layer in nm = 1e-9m % third column is the doping of the layer in 1e18 cm-3 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% substrate=InP; F0=0;%6e6;%0; % Electric field [Volt/meter] M=[ AlInAs 5 0 AlInAs 1 7 AlInAs 5 0 InGaAs 6 0 AlInAs 5 0 InGaAs 7 0 AlInAs 5 0 AlInAs 1 5 AlInAs 5 0 ]; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % substrate=GaAs; % Important for the Strain model (Si, GaAs, InP, InAs, GaSb) % F0=0;%6e6;%0; % Electric field [Volt/meter] % M=[ % GaAs 10 0.5 % InGaAs20 15 0 % GaAs 10 0.5 % ]; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % substrate=GaAs; % Important for the Strain model (Si, GaAs, InP, InAs, GaSb) % F0=0;%16e6;%0; % Electric field [Volt/meter] % M=[ % AlGaAs40 5 2 % GaAs 10 0 % AlGaAs40 5 2 % ]; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
find_Ef_f.m
.m
1,701
54
function[Ef,NN,roEf]=find_Ef_f(z,Ec,psic,E,ro,Ntot,T) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Constants %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% e = 1.602176487E-19; %% electron charge [C] kB = 1.3806488E-23; %% Boltzmann's constant [J/K] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% EE = repmat(E' ,[1 length(z)]); %%%%%%%%%%%%%%%%%%% Computes the Fermi level at any T %%%%%%%%%%%%%%%%%%%%%%%%%% if T==0 T=1e-10; end Ef=Ec(1); Fermi= 1./(1+exp((EE-Ef)/(kB*T/e))) ; FFermi=repmat(Fermi,[1 1 length(Ec)]); roEf = ro.*FFermi; NNz = squeeze(trapz(E,roEf,1)); NN = trapz(z,NNz.*abs(psic).^2 ,1) ; NtotX=sum(NN); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Now, it will try to get as close as posible to the real Ef with an % error of 0.1% by dichotomy ddE=0.01; % eV Ef1=Ef; Ef2=Ef1+ddE; while abs(NtotX - Ntot)/Ntot > 0.001 % find the Fermi level at any temperature if NtotX > Ntot Ef = Ef - abs(Ef1-Ef2)/2 ; Ef1 = Ef ; else Ef = Ef + abs(Ef1-Ef2)/2 ; Ef2 = Ef ; end Fermi = 1./(1+exp((EE-Ef)/(kB*T/e))) ; % Fermi Dirac distribution function FFermi = repmat(Fermi,[1 1 length(Ec)]); roEf = ro.*FFermi; NNz = squeeze(trapz(E,roEf,1)); NN = trapz(z,NNz.*abs(psic).^2 ,1) ; NtotX = sum(NN); end end
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
Schrod_Nbands_shoot_f.m
.m
3,601
116
function[E,psi]=Schrod_Nbands_shoot_f(z,V0,me,n,Evec,dE,precision) method=2; % method "2" is much more acurate than "1" but take also more time... e=min(V0); Emax=max(V0)+0.1; C=0; N=0; E=[]; psi=z*0+1; psi_old=psi; while e<Emax && length(E)<n C = C+1; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % In this block, it is scanning to find the mass at the right Energy, m(E) idx=[0 0]; epsi=1; while length(idx)>1 idx=find( abs( (Evec-e )) < epsi ) ; if isempty(idx) idx=IDX(1); else IDX=idx; epsi=epsi/2; end end Mass=me(idx,:); % [ e Mass(round(end/2)) ]; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% psi_old = psi; psi=Schroed1D_Euler_Eval(z,V0,Mass,e,method); if (sign( psi(end) ) ~= sign( psi_old(end) ) ) && C>1 % here, I catch a quantum state because the last point of psi change sign N=N+1; de=dE; while abs(de)>precision if sign( psi(end) ) ~= sign( psi_old(end) ) de = -de/2; end e=e+de; psi_old=psi; psi=Schroed1D_Euler_Eval(z,V0,Mass,e,method); end E(N,:)=e; PSI(:,N)= psi; C=0; end e=e+dE; end psi=PSI; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%% Normalization of the Wavefunction %%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% for i=1:N z=z'; psi(:,i)=psi(:,i)/sqrt(trapz(z,abs(psi(:,i)).^2)); end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function[psi]=Schroed1D_Euler_Eval(z,V0,Mass,ee,method) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% h=6.62606896E-34; %% Planck constant [J.s] hbar=h/(2*pi); e=1.602176487E-19; %% electron charge [C] m0=9.10938188E-31; %% electron mass [kg] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% dz=z(2)-z(1); dz=[ dz diff(z)]; y = [0 1]; if method==1 for i = 1:length(z) %% number of z steps to take dy(2) = -2*e/(hbar^2) * ( ee - V0(i) ) * y(1) ; %% Equation for dv/dz dy(1) = Mass(i)*m0*y(2); %% Equation for dx/dz y = y + dz(i)*dy ; %% integrate both equations with Euler psi(i)= y(1); end elseif method==2 for i = 1:length(z) %% number of z steps to take dy(2) = -2*e/(hbar^2) * ( ee - V0(i) ) * y(1) ; %% Equation for dv/dz dy(1) = Mass(i)*m0*y(2); %% Equation for dx/dz K = y + 0.5*dz(i)*dy; dK(2) = -2*e/(hbar^2) * ( ee - V0(i) ) * K(1) ; %% Equation for dv/dz dK(1) = Mass(i)*m0*K(2); %% Equation for dx/dz y = y + dz(i)*dK; psi(i)= y(1); end end end
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
PrintResults.m
.m
1,065
25
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Print Results %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %display('') display('===================================================') display('Intersubband Results:') display('===================================================') %display('') for i=1:length(Ec) for j=1:length(Ec) if j>i display(strcat(... 'e',num2str(i),'-e',num2str(j),' = ',num2str( Ec(j)-Ec(i),'%.3f' ),'eV; z'... ,num2str(i),'-',num2str(j),' = ',num2str( z_dipole_c(i,j)*1e9,'%.3f' ),'nm; f'... ,num2str(i),'-',num2str(j),' = ',num2str( f_dipole_c(i,j),'%.3f' ) ... ) ) end end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
ExtractParameters.m
.m
8,801
271
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%% Extract general parameters %%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % lattice parameter for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'a'); if idx==1 idx_a=i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Epsilon static for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'Epsi_stat'); if idx==1 idx_Epsi=i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % EP from the Kane model for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'EP_Kane'); if idx==1 idx_EP_K = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % EP from the Luttinger model % for i=1:length(DB.textdata(:,1)) % idx=strcmp(DB.textdata{i,1},'EP_Luttinger'); % if idx==1 % idx_EP_L = i-1; % % break % removing the break makes it slower but more compatible between Matlab and Octave % end % end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Energy level of the band 7c %for i=1:length(DB.textdata(:,1)) % idx=strcmp(DB.textdata{i,1},'Eg7c'); % if idx==1 % idx_Eg7c = i-1; % % break % removing the break makes it slower but more compatible between Matlab and Octave % end %end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Energy level of the band 8c %for i=1:length(DB.textdata(:,1)) % idx=strcmp(DB.textdata{i,1},'Eg8c'); % if idx==1 % idx_Eg8c = i-1; % % break % removing the break makes it slower but more compatible between Matlab and Octave % end %end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Energy level of the band 6v %for i=1:length(DB.textdata(:,1)) % idx=strcmp(DB.textdata{i,1},'Eg6v'); % if idx==1 % idx_Eg6v = i-1; % % break % removing the break makes it slower but more compatible between Matlab and Octave % end %end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Energy level of the band 6c (the band gap in a direct band gap semiconductor) for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'Eg6cG'); if idx==1 idx_Eg6c = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Bandgap Temperature dependency parameter "alpha" at the Gamma point for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'alphaG'); if idx==1 idx_alphaG = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Bandgap Temperature dependency parameter "beta" at the Gamma point for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'betaG'); if idx==1 idx_betaG = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % split-off band energy Delta-so for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'Dso'); if idx==1 idx_Dso = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Luttinger parameter for the electrons % for i=1:length(DB.textdata(:,1)) % idx=strcmp(DB.textdata{i,1},'F'); % if idx==1 % idx_F = i-1; % % break % removing the break makes it slower but more compatible between Matlab and Octave % end % end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Luttinger parameter gamma1 % for i=1:length(DB.textdata(:,1)) % idx=strcmp(DB.textdata{i,1},'gamma1'); % if idx==1 % idx_g1 = i-1; % % break % removing the break makes it slower but more compatible between Matlab and Octave % end % end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Luttinger parameter gamma2 % for i=1:length(DB.textdata(:,1)) % idx=strcmp(DB.textdata{i,1},'gamma2'); % if idx==1 % idx_g2 = i-1; % % break % removing the break makes it slower but more compatible between Matlab and Octave % end % end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Luttinger parameter gamma3 % for i=1:length(DB.textdata(:,1)) % idx=strcmp(DB.textdata{i,1},'gamma3'); % if idx==1 % idx_g3 = i-1; % % break % removing the break makes it slower but more compatible between Matlab and Octave % end % end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Valence band offset for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'VBO'); if idx==1 idx_VBO = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % mass for the 1band Schrodinger solver for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'me'); if idx==1 idx_me = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%% Strain parameters %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % ac Conduction band Strain parameter for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'ac'); if idx==1 idx_ac = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % av Valence band Strain parameter for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'av'); if idx==1 idx_av = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % bv Valence band Strain parameter for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'bv'); if idx==1 idx_bv = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % dv Valence band Strain parameter %for i=1:length(DB.textdata(:,1)) % idx=strcmp(DB.textdata{i,1},'dv'); % if idx==1 % idx_dv = i-1; % % break % removing the break makes it slower but more compatible between Matlab and Octave % end %end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % c11 Valence band Strain parameter for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'c11'); if idx==1 idx_c11 = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % c12 Valence band Strain parameter for i=1:length(DB.textdata(:,1)) idx=strcmp(DB.textdata{i,1},'c12'); if idx==1 idx_c12 = i-1; % break % removing the break makes it slower but more compatible between Matlab and Octave end end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % c44 Valence band Strain parameter %for i=1:length(DB.textdata(:,1)) % idx=strcmp(DB.textdata{i,1},'c44'); % if idx==1 % idx_c44 = i-1; % % break % removing the break makes it slower but more compatible between Matlab and Octave % end %end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
MATLAB
2D
LaurentNevou/Q_SchrodingerPoisson1D_CB
Schrod_2bands_Kane_f.m
.m
4,189
121
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%% Schrodinger solver on uniform grid with m(z,E)!!! %%%%%%%%%%%%%%% %%%%%%%%%%%%%%%% With the non-parabolic band 2x2k.p Kane model %%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function[Ec,psi_c]=Schrod_2bands_Kane_f(z,Vc,Eg,EP,Dso,n,ac,av,bv,exx,ezz) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Constants %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% h=6.62606896E-34; %% Planck constant J.s hbar=h/(2*pi); e=1.602176487E-19; %% charge de l electron Coulomb m0=9.10938188E-31; %% electron mass kg %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Nz=length(z); dz = z(2)-z(1); eyy = exx; DCBO = -abs(ac).*(exx+eyy+ezz) ; % shift of the CB due to strain DVBOLH = +abs(av).*(exx+eyy+ezz) + abs(bv).*(exx-ezz) ; % shift of the VB due to strain DVBOSO = +abs(av).*(exx+eyy+ezz) ; % shift of the VB due to strain Vc=Vc+DCBO; Vc(1)=5; Vc(end)=5; shift=min(Vc); Vc=Vc-shift; Vv=Vc-Eg+DVBOLH; Vso=Vc-Dso-Eg+DVBOSO; Vveff= (2*Vv+Vso)/3; % here is an effective valence band that make a ratio between LH and SO %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%% Building of the operators %%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%% First derivative %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %DZ1c = (0.5)*diag(ones(1,Nz-1),+1) + (-0.5)*diag(ones(1,Nz-1),-1) ; DZ1b = (1)*diag(ones(1,Nz) ,0 ) + (-1)*diag(ones(1,Nz-1),-1) ; DZ1f = (1)*diag(ones(1,Nz-1),+1) + (-1)*diag(ones(1,Nz) ,0 ) ; %DZ1c=DZ1c/dz; DZ1b=DZ1b/dz; DZ1f=DZ1f/dz; %%%%%%%%%%%%%%%%%%%%%%%%%%%% Second derivative %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% DZ2 =(-2)*diag(ones(1,Nz)) + (1)*diag(ones(1,Nz-1),-1) + (1)*diag(ones(1,Nz-1),1); DZ2=DZ2/dz^2; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%% Building of the Hamiltonien %%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Vc = [ (Vc(1:end-1) + Vc(2:end)) / 2 Vc(end) ]; Vveff = [ (Vveff(1:end-1) + Vveff(2:end))/ 2 Vveff(end) ]; EP = [ (EP(1:end-1) + EP(2:end)) / 2 EP(end) ]; H0=(-(hbar^2)/(2*m0)) * DZ2 ; H11 = H0 + diag(Vc*e) ; H22 = -H0 + diag( (Vveff)*e ) ; % Xunpeng Ma et al. JAP, 114, 063101 (2013) % "Two-band finite difference method for the bandstructure calculation with nonparabolicity effects in quantum cascade lasers" % ==> It s seems to be by far the most accurate method H12 = +1*hbar/sqrt(2*m0) * ( diag(sqrt(EP*e),0) + diag(sqrt(EP(1:end-1)*e),-1) ) .* DZ1b ; H21 = -1*hbar/sqrt(2*m0) * ( diag(sqrt(EP*e),0) + diag(sqrt(EP(1:end-1)*e),+1) ) .* DZ1f ; H2x2=[ H11 H12 H21 H22 ]; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%% Diagonalisation of the Hamiltonien %%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% H2x2=sparse(H2x2); [psi_2x2,Energy] = eigs(H2x2,n,'SM'); E_2x2 = diag(Energy)/e ; psi_c=[]; Ec=[]; for i=1:n if E_2x2(i) > min(Vc) Ec=[Ec abs(E_2x2(i))]; psi_c = [psi_c psi_2x2(1:length(z),i) ]; end end Ec=Ec'+shift; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%% Normalization of the Wavefunction %%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% for i=1:length(Ec) psi_c(:,i)=psi_c(:,i)/sqrt(trapz(z',abs(psi_c(:,i)).^2)); end %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % here is a small patch due to differences between Octave and Matlab % Matlab order the eigen values while Octave reverse it if length(Ec)>1 if Ec(1)>Ec(2) psi_c=psi_c(:,end:-1:1); Ec=Ec(end:-1:1); end end end
MATLAB
2D
obaica/vasp_wavecar_parity
main.f90
.f90
587
19
program main implicit none integer :: ierr ierr=0 open(unit=19, file='GCOEFF.txt', status='old', action='read', iostat=ierr) close(unit=19) if ( ierr.eq.0 ) then write(*,*) "'GCOEFF.txt' file exits, now calc parity." call parity write(*,*) "**************End of parity calculation!*********************" else write(*,*) "'GCOEFF.txt' file dosn't exit, now read WAVECAR." call wavetrans call parity write(*,*) "**************End of parity calculation!*********************" end if end program main
Fortran
2D
obaica/vasp_wavecar_parity
wavetrans.f90
.f90
10,118
310
!!$************************* WaveTrans !************************************ !!$* !!$* input the WAVECAR file in binary format from VASP, and output !!$* in text format a file GCOEFF of G values and corresponding plane !!$* wave coefficients, together with energy eigenvalues and !occupations !!$* !!$* Compile with gfortran or ifort. Flag "-assume byterecl" is !required !!$* for ifort. !!$* !!$* version 2.0 - July 5, 2012 - R. M. Feenstra and M. Widom !!$* version 2.1 - Sept 30, 2012 - changed estimator for max. no. of !!$* plane waves !!$* version 1.3 - Sept 17, 2014 - updated 'c' value !!$* !!$* format of GCOEFF file: !!$* no. wavevector values (integer) !!$* no. bands (integer) !!$* real space lattice vector (a1) x,y,z coefficients (real) !!$* real space lattice vector (a2) x,y,z coefficients (real) !!$* real space lattice vector (a3) x,y,z coefficients (real) !!$* recip. lattice vector 1 (b1) x,y,z coefficients (real) !!$* recip. lattice vector 2 (b2) x,y,z coefficients (real) !!$* recip. lattice vector 3 (b3) x,y,z coefficients (real) !!$* loop over spins !!$* loop through wavevector values: !!$* wavevector x,y,z components (real) !!$* energy (complex), occupation (real) !!$* loop through bands: !!$* index of band (integer), no. plane waves (integer) !!$* loop through plane waves: !!$* ig1,ig2,ig3 values (integer) and coefficient !(complex) !!$* end loop through plane waves !!$* end loop through bands !!$* end loop through wavevector values !!$* end loop over spins !!$* !!$* the x,y,z components of each G value are given in terms of the !!$* ig1,ig2,ig3 values and the components of the recip. lattice !vectors !!$* according to: !!$* ig1*b1_x + ig2*b2_x + ig3*b3_x, !!$* ig1*b1_y + ig2*b2_y + ig3*b3_y, and !!$* ig1*b1_z + ig2*b2_z + ig3*b3_z, respectively !!$* !!$* note that the energy eigenvalues are complex, as provided in the !!$* WAVECAR file, but the imaginary part is zero (at least for all !cases !!$* investigated thus far) !!$* subroutine wavetrans implicit real*8 (a-h, o-z) character :: soc complex*8, allocatable :: coeff(:,:) complex*16, allocatable :: cener(:) real*8, allocatable :: occ(:) integer, allocatable :: igall(:,:) dimension a1(3),a2(3),a3(3),b1(3),b2(3),b3(3),vtmp(3),sumkg(3),wk(3) !!$* constant 'c' below is 2m/hbar**2 in units of 1/eV Ang^2 (value is !!$* adjusted in final decimal places to agree with VASP value; !!$* checks for discrepancy of any results between this and VASP !values) data c/0.262465831d0/ !!$* data c/0.26246582250210965422d0/ pi=4.*atan(1.) !!$* input soc='n' nrecl=24 do write(*,*)"Calculation with SoC ? 'y/n'" read(*,*) soc if ( soc.eq.'y' ) then write(*,*)"Calculation with SoC!" exit else if ( soc.eq.'n' ) then write(*,*)"Calculation without SoC!" exit else write(*,*)"Input error, y or n" end if end do write(*,*) open(unit=10,file='WAVECAR',access='direct',recl=nrecl, & iostat=iost,status='old') if (iost.ne.0) write(6,*) 'open error - iostat =',iost read(unit=10,rec=1) xnrecl,xnspin,xnprec close(unit=10) nrecl=nint(xnrecl) nspin=nint(xnspin) nprec=nint(xnprec) if(nprec.eq.45210) then write(6,*) '*** error - WAVECAR_double requires complex*16' stop endif write(6,*) write(6,*) 'record length =',nrecl,' spins =',nspin, & 'prec flag ',nprec open(unit=10,file='WAVECAR',access='direct',recl=nrecl, & iostat=iost,status='old') if (iost.ne.0) write(6,*) 'open error - iostat =',iost open(unit=11,file='GCOEFF.txt') read(unit=10,rec=2) xnwk,xnband,ecut,(a1(j),j=1,3),(a2(j),j=1,3), & (a3(j),j=1,3) nwk=nint(xnwk) if (nwk.eq.0) write(*,*) 'no kpoints in WAVECAR or reading error' nband=nint(xnband) allocate(occ(nband)) allocate(cener(nband)) write(6,*) 'no. k points =',nwk write(6,*) 'no. bands =',nband write(6,*) 'max. energy =',sngl(ecut) write(6,*) 'real space lattice vectors:' write(6,*) 'a1 =',(sngl(a1(j)),j=1,3) write(6,*) 'a2 =',(sngl(a2(j)),j=1,3) write(6,*) 'a3 =',(sngl(a3(j)),j=1,3) write(6,*) ' ' write(11,*) nspin write(11,*) nwk write(11,*) nband !!$* compute reciprocal properties write(6,*) ' ' call vcross(vtmp,a2,a3) Vcell=a1(1)*vtmp(1)+a1(2)*vtmp(2)+a1(3)*vtmp(3) write(11,*) Vcell write(6,*) 'volume unit cell =',sngl(Vcell) call vcross(b1,a2,a3) call vcross(b2,a3,a1) call vcross(b3,a1,a2) do j=1,3 b1(j)=2.*pi*b1(j)/Vcell b2(j)=2.*pi*b2(j)/Vcell b3(j)=2.*pi*b3(j)/Vcell enddo b1mag=dsqrt(b1(1)**2+b1(2)**2+b1(3)**2) b2mag=dsqrt(b2(1)**2+b2(2)**2+b2(3)**2) b3mag=dsqrt(b3(1)**2+b3(2)**2+b3(3)**2) write(6,*) 'reciprocal lattice vectors:' write(6,*) 'b1 =',(sngl(b1(j)),j=1,3) write(6,*) 'b2 =',(sngl(b2(j)),j=1,3) write(6,*) 'b3 =',(sngl(b3(j)),j=1,3) write(6,*) 'reciprocal lattice vector magnitudes:' write(6,*) sngl(b1mag),sngl(b2mag),sngl(b3mag) write(6,*) ' ' write(11,*) (sngl(a1(j)),j=1,3) write(11,*) (sngl(a2(j)),j=1,3) write(11,*) (sngl(a3(j)),j=1,3) write(11,*) (sngl(b1(j)),j=1,3) write(11,*) (sngl(b2(j)),j=1,3) write(11,*) (sngl(b3(j)),j=1,3) phi12=acos((b1(1)*b2(1)+b1(2)*b2(2)+b1(3)*b2(3))/(b1mag*b2mag)) call vcross(vtmp,b1,b2) vmag=dsqrt(vtmp(1)**2+vtmp(2)**2+vtmp(3)**2) sinphi123=(b3(1)*vtmp(1)+b3(2)*vtmp(2)+b3(3)*vtmp(3))/(vmag*b3mag) nb1maxA=(dsqrt(ecut*c)/(b1mag*abs(sin(phi12))))+1 nb2maxA=(dsqrt(ecut*c)/(b2mag*abs(sin(phi12))))+1 nb3maxA=(dsqrt(ecut*c)/(b3mag*abs(sinphi123)))+1 npmaxA=nint(4.*pi*nb1maxA*nb2maxA*nb3maxA/3.) phi13=acos((b1(1)*b3(1)+b1(2)*b3(2)+b1(3)*b3(3))/(b1mag*b3mag)) call vcross(vtmp,b1,b3) vmag=dsqrt(vtmp(1)**2+vtmp(2)**2+vtmp(3)**2) sinphi123=(b2(1)*vtmp(1)+b2(2)*vtmp(2)+b2(3)*vtmp(3))/(vmag*b2mag) phi123=abs(asin(sinphi123)) nb1maxB=(dsqrt(ecut*c)/(b1mag*abs(sin(phi13))))+1 nb2maxB=(dsqrt(ecut*c)/(b2mag*abs(sinphi123)))+1 nb3maxB=(dsqrt(ecut*c)/(b3mag*abs(sin(phi13))))+1 npmaxB=nint(4.*pi*nb1maxB*nb2maxB*nb3maxB/3.) phi23=acos((b2(1)*b3(1)+b2(2)*b3(2)+b2(3)*b3(3))/(b2mag*b3mag)) call vcross(vtmp,b2,b3) vmag=dsqrt(vtmp(1)**2+vtmp(2)**2+vtmp(3)**2) sinphi123=(b1(1)*vtmp(1)+b1(2)*vtmp(2)+b1(3)*vtmp(3))/(vmag*b1mag) phi123=abs(asin(sinphi123)) nb1maxC=(dsqrt(ecut*c)/(b1mag*abs(sinphi123)))+1 nb2maxC=(dsqrt(ecut*c)/(b2mag*abs(sin(phi23))))+1 nb3maxC=(dsqrt(ecut*c)/(b3mag*abs(sin(phi23))))+1 npmaxC=nint(4.*pi*nb1maxC*nb2maxC*nb3maxC/3.) nb1max=max0(nb1maxA,nb1maxB,nb1maxC) nb2max=max0(nb2maxA,nb2maxB,nb2maxC) nb3max=max0(nb3maxA,nb3maxB,nb3maxC) if ( soc.eq.'y') then npmax=2*min0(npmaxA,npmaxB,npmaxC) else if ( soc.eq.'n' ) then npmax=min0(npmaxA,npmaxB,npmaxC) endif write(6,*) 'max. no. G values; 1,2,3 =',nb1max,nb2max,nb3max write(6,*) ' ' allocate (igall(3,npmax)) write(6,*) 'estimated max. no. plane waves =',npmax write(11, *) npmax allocate (coeff(npmax,nband)) !!$* Begin loops over spin, k-points and bands irec=2 do isp=1,nspin write(*,*) ' ' write(*,*) '******' write(*,*) 'reading spin ',isp do iwk=1,nwk irec=irec+1 read(unit=10,rec=irec) xnplane,(wk(i),i=1,3), & (cener(iband),occ(iband),iband=1,nband) nplane=nint(xnplane) write(6,*) 'k point #',iwk,' input no. of plane waves =', & nplane write(6,*) 'k value =',(sngl(wk(j)),j=1,3) write(11,*) (sngl(wk(j)),j=1,3) !!$* Calculate plane waves ncnt=0 do ig3=0,2*nb3max ig3p=ig3 if (ig3.gt.nb3max) ig3p=ig3-2*nb3max-1 do ig2=0,2*nb2max ig2p=ig2 if (ig2.gt.nb2max) ig2p=ig2-2*nb2max-1 do ig1=0,2*nb1max ig1p=ig1 if (ig1.gt.nb1max) ig1p=ig1-2*nb1max-1 do j=1,3 sumkg(j)=(wk(1)+ig1p)*b1(j)+ & (wk(2)+ig2p)*b2(j)+(wk(3)+ig3p)*b3(j) enddo gtot=sqrt(sumkg(1)**2+sumkg(2)**2+sumkg(3)**2) etot=gtot**2/c if (etot.lt.ecut) then ncnt=ncnt+1 igall(1,ncnt)=ig1p igall(2,ncnt)=ig2p igall(3,ncnt)=ig3p end if enddo enddo enddo if ( soc.eq.'y') then if (2*ncnt.ne.nplane) then write(6,*) 'ncnt=',ncnt, '// nplane=',nplane, 'npmax=', npmax write(6,*) '*** error - computed no. != input no.' stop endif else if ( soc.eq.'n' ) then if (ncnt.ne.nplane) then write(6,*) 'ncnt=',ncnt, '// nplane=',nplane, 'npmax=', npmax write(6,*) '*** error - computed no. != input no.' stop endif endif if (ncnt.gt.npmax) then write(6,*) '*** error - plane wave count exceeds estimate' stop endif do iband=1,nband irec=irec+1 read(unit=10,rec=irec) (coeff(iplane,iband), & iplane=1,nplane) enddo !!$* output G values and coefficients 560 format('( ',g14.6,' , ',g14.6,' ) ',g14.6) 570 format(3i6,' ( ',g14.6,' , ',g14.6,' )') if ( soc.eq.'y' ) then do iband=1,nband write(11,*) iband,nplane write(11,560) cener(iband),occ(iband) do iplane=1,nplane write(11,570) (igall(j,mod(iplane-1, ncnt)+1),j=1,3), & coeff(iplane,iband) enddo enddo else do iband=1,nband write(11,*) iband,nplane write(11,560) cener(iband),occ(iband) do iplane=1,nplane write(11,570) (igall(j,iplane),j=1,3), & coeff(iplane,iband) enddo enddo end if enddo enddo end subroutine wavetrans !!$* !!$* routine for computing vector cross-product !!$* subroutine vcross(a,b,c) implicit real*8(a-h,o-z) dimension a(3),b(3),c(3) a(1)=b(2)*c(3)-b(3)*c(2) a(2)=b(3)*c(1)-b(1)*c(3) a(3)=b(1)*c(2)-b(2)*c(1) return end subroutine vcross
Fortran
2D
obaica/vasp_wavecar_parity
parity.f90
.f90
5,331
172
subroutine parity implicit none complex, parameter :: zi=(0.000, 0.000) integer :: i, j, l, n, m real:: V complex, allocatable :: coeff(:, :) integer :: nspin, nkpt, nband, npmax integer, allocatable :: Gvector(:, :, :) real :: a1(3),a2(3),a3(3), b1(3),b2(3),b3(3) real :: kcurrent(3) real :: ktarget(3) integer :: bandindex, nplane complex :: energy real :: occ integer :: inot,rpt real :: r0(3), is(3) complex :: psi1, psi2 real :: pindex complex, allocatable :: parityresult(:,:,:) real :: diff complex :: mp character :: pos a1=0.00;a2=0.00;a3=0.00 b1=0.00;b2=0.00;b3=0.00 rpt=5 is=[0.00, 0.00, 0.00] open(unit=10, file='GCOEFF.txt', status='old') write(*,*) "Specified the target kpoint (kx, ky, kz):" read(*, *) ktarget(1), ktarget(2), ktarget(3) write(*,*) "Specified the Inversion Centre (x, y, z):" read(*,*) is(1), is(2), is(3) read(10, *) nspin read(10, *) nkpt read(10, *) nband read(10, *) V read(10, *) a1(1), a1(2), a1(3) read(10, *) a2(1), a2(2), a2(3) read(10, *) a3(1), a3(2), a3(3) read(10, *) b1(1), b1(2), b1(3) read(10, *) b2(1), b2(2), b2(3) read(10, *) b3(1), b3(2), b3(3) read(10, *) npmax allocate(coeff(nband, npmax)) allocate(Gvector(nband, npmax, 3)) allocate(parityresult(5,nband,2)) Gvector=0.00 coeff=0.00 pindex=0 do i=1, nkpt inot=0 ! assume current k point not in target k list read(10, *) kcurrent(1), kcurrent(2), kcurrent(3) do j=1, nband read(10, *) bandindex, nplane read(10, *) energy, occ do l=1, nplane read(10, *) Gvector(bandindex,l,1),Gvector(bandindex,l,2), Gvector(bandindex, l, 3), coeff(j, l) ! write(*, *) Gvector(bandindex, l, :) end do ! l=1, nplane end do ! j=1, nband !call inlist(kcurrent, ktarget, nktarget, inot) diff = sqrt((kcurrent(1)-ktarget(1))**2+(kcurrent(2)-ktarget(2))**2& +(kcurrent(3)-ktarget(3))**2) if ( diff<=1d-5 ) then ! if the current kpt is target kpt write(*, *) 'current k point:', kcurrent write(*, *) 'target k point:', ktarget write(*,*) "#band index, parity, wf(+r)/wf(-r)" do l=1, nband do m=1, rpt call rvector(r0) call wavefun(Gvector(l,1:nplane,:),coeff(l,1:nplane),nplane,kcurrent, r0, is, V, psi1, psi2) parityresult(m, l, 1)= psi1 parityresult(m, l, 2)= psi2 end do end do ! l=1, nband do l=1, nband mp=0.00 do m=1,rpt !if (mod(l,2).eq.0) cycle mp=mp+parityresult(m,l,1)/parityresult(m,l, 2) end do mp = mp/dble(rpt) !take 5 points to avoid accident fault if (real(mp) < -0.99 .and. real(mp) > -1.01) then pos="-" else if ( real(mp)>0.99 .and. real(mp)<1.01) then pos="+" end if write(*,770)l, pos, mp 770 format (I10, A4, ' (', ES15.6, ', i', ES15.6, ')') end do exit end if ! if kcurrent in ktarget end do ! i=1, npkt close(10) write(*,*)"### IF the parity of bands are not +/- or not showing " write(*,*)"### or wf(+)/wf(-) are not around +/-(1., i 0.) then " write(*,*)"### you need reconsider the INVERSION CENTRE coordinate!" end subroutine parity subroutine wavefun(G, C, npl, k ,r0, is, v, psi1, psi2) ! this subroutine aims at construct wavefuncion giving coefficients implicit none integer :: i complex, parameter :: pzi=2.0*3.141592654*(0.00,1.0000) complex, parameter :: zi=(0.00, 1.00) integer, intent(in) :: npl integer, intent(in) :: G(npl, 3) complex, intent(in) :: C(npl) real, intent(in) :: k(3), r0(3), v, is(3) real :: r1(3) complex, intent(out) :: psi1, psi2 real :: ck(3) real :: dotpd0, dotpd1 psi1=(0.000,0.000) psi2=(0.000,0.000) r1=2.0*is-r0 do i=1,npl ck = G(i, :) + k dotpd0 = ck(1)*r0(1) + ck(2)*r0(2) + ck(3)*r0(3) dotpd1 = ck(1)*r1(1) + ck(2)*r1(2) + ck(3)*r1(3) psi1 = psi1 + C(i)*exp(pzi*dotpd0)/sqrt(v) psi2 = psi2 + C(i)*exp(pzi*dotpd1)/sqrt(v) end do end subroutine wavefun subroutine inlist(e, list, l, r) ! !PURPUSE: check if element E in LIST or not !PARAMETERS: ! e: element ! list: target list ! l: lenth of target list ! r: result, r=1 for e IN list ! implicit none integer, intent(in) :: l real,intent(in):: e(3), list(l, 3) integer, intent(inout) :: r integer :: n real:: tol, diff tol=1d-6 diff=1d2 do n=1, l diff=sqrt((e(1)-list(n,1))**2+(e(2)-list(n,2))**2+(e(3)-list(n,3))**2) if ( diff.lt.tol ) then r=1 exit end if ! diff end do ! n=1, l end subroutine inlist subroutine rvector(r) implicit none real, intent(out) :: r(3) integer :: i real :: x call random_seed() do i=1,3 call random_number(x) r(i)=2.0*x-1.0 end do end subroutine rvector
Fortran
2D
poreathon/poreathon
getdata.sh
.sh
327
10
#!/bin/bash cd input/nanopore wget http://pathogenomics.bham.ac.uk/filedist/nanopore/Ecoli_R7_2D.fastq wget http://pathogenomics.bham.ac.uk/filedist/nanopore/Ecoli_R7_2D.fasta wget http://pathogenomics.bham.ac.uk/filedist/nanopore/Ecoli_R73_2D.fasta wget http://pathogenomics.bham.ac.uk/filedist/nanopore/Ecoli_R73_2D.fastq
Shell
2D
poreathon/poreathon
makemakefile.py
.py
794
23
import glob import yaml pipelines = yaml.load(file('pipelines.yaml')) recipes = glob.glob('recipes/*.yaml') recipes = dict([(r[8:-5], yaml.load(file(r))) for r in recipes]) for pipeline, inputs in pipelines.iteritems(): for input_file in inputs: for recipe_name, recipe_info in recipes.iteritems(): if recipe_info['type'] == pipeline: if recipe_info['input'] == 'fasta': suffix = 'fasta' elif recipe_info['input'] == 'fastq': suffix = 'fastq' else: raise SystemError("Wrong suffix name in YAML file %s" % (recipe_name)) print "bpipe run -r -f %s_%s.html -d output/%s/%s -p ENTRYPOINT=recipes/%s.pipe @params.txt pipelines/%s.pipe input/nanopore/%s.%s" % (recipe_name, input_file, pipeline, recipe_name, recipe_name, pipeline, input_file, suffix)
Python
2D
poreathon/poreathon
PIPELINES.md
.md
1,322
67
# Proposed Pipelines ## Alignment The alignment pipeline aligns FASTA or FASTQ or FAST5 files to a reference. Input: FASTA/FASTQ/FAST5 file & reference Output: Sorted BAM file Tests: - # mapped/ # unmapped - median/mean coverage - time to run - histogram of alignment lengths - coverage plots ## Alignment corrector This pipeline takes an alignment and post-processes it, e.g. with the aim of improving SNP or indel calling sensitivity/specificity. Input: Sorted BAM file Output: Sorted BAM file Tests: - ?? ## Consensus calling (normal reference) This pipeline takes results from alignment or alignment corrector and produces a consensus output. Input: FASTA file & reference Output: VCF file Tests: - number of bases correct - number of bases called - number of bases incorrect (SNPs, indels) ## Consensus calling (mutated references) The reference is mutated, e.g. 0.1%, 1%, 5% nucleotide divergence and a consensus is produced. Tests: - truth table - ROC curves ## Hybrid assembly De novo assembly using Illumina reads as guide. Tests: - QUAST output ## Nanopore-only assembly Nanopore only assembly. ## Read correction by Illumina Reads are corrected by Illumian reads. ## Read correction de novo Reads are corrected, e.g. by a statistical model or by FAST5 squiggles.
Markdown
2D
poreathon/poreathon
analysis/count-errors.py
.py
2,263
83
import sys import pysam from collections import Counter # http://pysam.readthedocs.org/en/latest/api.html#pysam.AlignedRead.cigar MATCH = 0 # M INS = 1 # I DEL = 2 # D SKIP = 3 # N SOFT = 4 # S HARD = 5 # H PAD = 6 # P EQUAL = 7 # = DIFF = 9 # X def cigar_profile(cigar_tuples): """ Return a dictionary that tabulates the total number of bases associated with each CIGAR operation. cigar_tuples is a list of (op, length) tuples. """ cigar_prof = Counter() for cigar_tuple in cigar_tuples: cigar_prof[cigar_tuple[0]] += cigar_tuple[1] return cigar_prof def get_total_differences(cigar_prof): """ return the total number of get_total_differences in the alignment between the query and the reference. (mismatches + insertions + deletions) """ return cigar_prof[DIFF] + cigar_prof[INS] + cigar_prof[DEL] def get_total_unaligned(cigar_prof): """ return the total number unaligned bases (hard or softclips.) """ return cigar_prof[HARD] + cigar_prof[SOFT] # Pass 1: # iterate through each BAM alignment # and store the best alignment for the read # best_align: # key is query name # val is tuple of (alignment length, cigar_prof) best_align = {} bam = pysam.Samfile(sys.argv[1]) for read in bam: cigar_prof = cigar_profile(read.cigar) if read.qname not in best_align: best_align[read.qname] = (read.alen, read.inferred_length, cigar_prof) elif read.qname in best_align \ and read.alen > best_align[read.qname][0]: best_align[read.qname] = (read.alen, read.inferred_length, cigar_prof) # Pass 2: # Report the alignment and error profile for each read's best alignment print '\t'.join(['query', 'read_type', 'read_len', 'align_len', 'unalign_len', 'matches', 'mismatches', 'insertions', 'deletions', 'tot_errors']) for query in best_align: read_type = query.split('_')[4] alen = best_align[query][0] inferred_length = best_align[query][1] cigar_prof = best_align[query][2] total_errors = get_total_differences(cigar_prof) unaligned_len = get_total_unaligned(cigar_prof) print '\t'.join(str(s) for s in [query, read_type, inferred_length, alen, unaligned_len, \ cigar_prof[EQUAL], \ cigar_prof[DIFF], \ cigar_prof[INS], \ cigar_prof[DEL], \ total_errors])
Python
2D
poreathon/poreathon
analysis/cov_histogram.R
.R
287
7
#!/usr/bin/Rscript library(ggplot2) args=commandArgs(TRUE) cov=read.table(args[1], sep="\t", header=F) p=ggplot(cov, aes(x=V2, y=V3)) + geom_bar(stat="identity") + scale_x_continuous("Coverage depth") + scale_y_continuous("Number of bases") + theme_bw() ggsave(filename=args[2], plot=p)
R
2D
poreathon/poreathon
analysis/aln_histogram.R
.R
342
9
#!/usr/bin/Rscript library(ggplot2) args=commandArgs(TRUE) cov=read.table(args[1], sep="\t", header=T, stringsAsFactors=T) cov=subset(cov, align_len != 'None') cov=cbind(cov,"frac" = as.numeric(cov$align_len) / as.numeric(cov$read_len)) p=ggplot(cov, aes(x=frac)) + geom_histogram() + xlim(0, 2) + theme_bw() ggsave(filename=args[2], plot=p)
R
2D
poreathon/poreathon
analysis/expand-cigar.py
.py
3,347
114
import sys import pysam import argparse """Author: Aaron R. Quinlan, December 2014""" # http://pysam.readthedocs.org/en/latest/api.html#pysam.AlignedRead.cigar MATCH = 0 # M INS = 1 # I DEL = 2 # D SKIP = 3 # N SOFT = 4 # S HARD = 5 # H PAD = 6 # P EQUAL = 7 # = DIFF = 8 # X def get_chrom(fasta_fh, chrom): """ Return the chromosome sequence """ return fasta_fh.fetch(chrom) def expand_match(qry_seq, ref_seq): """ Return an expanded list of CIGAR ops based upon nuceltotide matches (=, or 7) and mismatches (X, or 8) """ prev_op = None curr_op = None length = 1 for idx, q_nucl in enumerate(qry_seq): if q_nucl == ref_seq[idx]: curr_op = 7 # EQUAL (=) else: curr_op = 8 # DIFF (X) if curr_op == prev_op: length += 1 elif prev_op is not None: yield (prev_op, length) length = 1 prev_op = curr_op yield (curr_op, length) def main(args): bam = pysam.AlignmentFile(args.bam) fa = pysam.FastaFile(args.fasta) outfile = pysam.AlignmentFile("-", "wb", template=bam) prev_chrom_id = None curr_chrom_id = None curr_chrom_seq = None for read in bam: curr_chrom_id = read.reference_id # load the current chromosome into memory if curr_chrom_id != prev_chrom_id: curr_chrom_seq = get_chrom(fa, bam.getrname(curr_chrom_id)) # iterate through the existing CIGAR # and replace M ops with expanded X and = ops. ref_pos = read.reference_start qry_pos = 0 new_cigar = [] for cigar_tuple in read.cigar: op = cigar_tuple[0] op_len = cigar_tuple[1] if op == EQUAL or op == DIFF or op == MATCH: if op == MATCH: qry_seq = read.query_sequence[qry_pos:qry_pos + op_len] ref_seq = curr_chrom_seq[ref_pos:ref_pos + op_len] if qry_seq == ref_seq: new_cigar.append((7, len(qry_seq))) # EQUAL (=) else: # expand the M CIGAR op into X and = ops. for new_cigar_tuple in expand_match(qry_seq, ref_seq): new_cigar.append(new_cigar_tuple) else: new_cigar.append(cigar_tuple) ref_pos += op_len qry_pos += op_len elif op == DEL or op == SKIP: ref_pos += op_len new_cigar.append(cigar_tuple) elif op == INS or op == SOFT: qry_pos += op_len new_cigar.append(cigar_tuple) elif op == HARD: new_cigar.append(cigar_tuple) # replace the old CIGAR and write updated record. read.cigar=new_cigar outfile.write(read) prev_chrom_id=curr_chrom_id outfile.close() if __name__ == "__main__": parser=argparse.ArgumentParser(prog='expand_cigar.py') parser.add_argument('--bam', dest='bam', metavar='STRING', help='The sorted BAM file whose CIGARs you wish expand.') parser.add_argument('--fasta', dest='fasta', metavar='STRING', help='The reference genome used to create the BAM.') args=parser.parse_args() main(args)
Python
2D
poreathon/poreathon
analysis/basic-stats.py
.py
500
26
import sys import pysam from collections import Counter import numpy def stats(fn): unaligned = 0 best_align = {} bam = pysam.Samfile(fn) for read in bam: if read.is_unmapped: unaligned += 1 continue if read.qname not in best_align: best_align[read.qname] = read.alen print """- Filename: %s Mean_Alignment_Length: %d Unaligned_Reads: %d Aligned_Reads: %d""" % (fn, numpy.mean(best_align.values()), unaligned, len(best_align.values())) for fn in sys.argv[1:]: stats(fn)
Python
2D
poreathon/poreathon
analysis/collate_alignment_stats.py
.py
1,845
51
import glob import os import yaml import sys import pysam from collections import Counter from Bio import SeqIO import numpy def count_sequences(input_file, typ): return len([s for s in SeqIO.parse(open(input_file), typ)]) def dump_pipeline(input_file, recipe_name, recipe_info): input_lines = count_sequences("input/nanopore/%s.%s" % (input_file, recipe_info['input']), recipe_info['input']) ## this is all super hacky right now and will break eventually filenames = glob.glob("output/alignment/%s/%s.*.bam" % (recipe_name, input_file)) unaligned = 0 best_align = {} bam = pysam.Samfile(filenames[0]) for read in bam: if read.is_unmapped: unaligned += 1 continue if read.qname not in best_align: best_align[read.qname] = read.alen output = recipe_info output['Recipe'] = recipe_name output['Input'] = input_file output['Mean_Alignment_Length'] = int(numpy.mean(best_align.values())) output['Aligned_Reads'] = len(best_align.values()) output['Unaligned_Reads'] = input_lines - output['Aligned_Reads'] output['Percent_Aligned'] = "%.2f" % (100.0 / input_lines * output['Aligned_Reads']) print "- " + yaml.dump(output) filenames = glob.glob("output/alignment/%s/%s*.draw_histo.png" % (recipe_name, input_file)) os.system("cp %s ../poreathon.github.io/output/alignment/%s_%s.histogram.png" % (filenames[0], recipe_name, input_file)) pipelines = yaml.load(file('pipelines.yaml')) recipes = glob.glob('recipes/*.yaml') recipes = dict([(r[8:-5], yaml.load(file(r))) for r in recipes]) for input_file in pipelines['alignment']: for recipe_name, recipe_info in recipes.iteritems(): if recipe_info['type'] == 'alignment': dump_pipeline(input_file, recipe_name, recipe_info)
Python
2D
lhalloran/ERTplot
ERTplot.py
.py
2,432
66
""" ERTplot.py Landon Halloran 29-Nov-2016 Script to read in 2D electrical resistivity tomography (ERT) data (i.e., inversion output from RES2DINV) and topography data and plot it with much more user control than RES2DINV allows. Data must first be exported to .xls/.xlsx format (see example file). """ import matplotlib as mpl import matplotlib.pyplot as plt import numpy as np import pandas as pd ###### User must define these: ###### filein = "example/Resistivity_and_Topography_Data.xlsx" # where the data is located ncolours=12 # number of colours for plots colourscheme='viridis' # for others see https://matplotlib.org/examples/color/colormaps_reference.html linekeys=['LINE_A','LINE_B'] # labels of tabs of ERT data topokeys=['LINE_A_TOPO','LINE_B_TOPO'] # labels of tabs of topography data (e.g. from GNSS survey) ##################################### nsubfigs=np.size(linekeys) xl_file = pd.ExcelFile(filein) ERT={} for j in xl_file.sheet_names: ERT[j]=xl_file.parse(j) allrhos=[] for i in linekeys: rho=ERT[i]['Res'].values allrhos.append(rho) allrhos = np.concatenate(allrhos) clevels = np.logspace(np.log10(np.min(allrhos)),np.log10(np.max(allrhos)),num=ncolours,base=10) fig, axes = plt.subplots(nrows=nsubfigs, ncols=1,sharex=False) # turn on sharex to share x! i=0 for ax in axes.flat: keyhere=linekeys[i] x=ERT[keyhere]['X'] z=ERT[keyhere]['Z'] rho=ERT[keyhere]['Res'] triang = mpl.tri.Triangulation(x, z) mask = mpl.tri.TriAnalyzer(triang).get_flat_tri_mask(.1) triang.set_mask(mask) topokeyhere=topokeys[i] xt=ERT[topokeyhere]['X'] zt=ERT[topokeyhere]['Z'] #plt.tricontourf(triang,rho,levels=clevels, cmap=colourscheme) #cc=ax.tricontourf(triang,rho,levels=clevels, cmap=colourscheme) cc=ax.tricontourf(triang,rho,levels=clevels, norm=mpl.colors.LogNorm(vmin=allrhos.min(), vmax=allrhos.max()), cmap=colourscheme) ax.axis('equal') # make axes equal (i.e. 1m x = 1m y) topokeyhere=topokeys[i] ax.plot(xt,zt,'k') i=i+1 fig.text(0.5, 0.04, 'X (m)', ha='center') fig.text(0.04, 0.5, 'Z (m)', va='center', rotation='vertical') fig.suptitle('Resistivity ($\Omega\cdot$m)') clabels=[] for c in clevels: clabels.append('%d' % c) # label all levels with no-decimal formatting thecbar=fig.colorbar(cc, ax=axes,ticks=clevels) thecbar.ax.set_yticklabels(clabels)
Python
2D
pl992/VirusPropagator
VirusPropagator.py
.py
3,859
134
import numpy as np import os import matplotlib.pyplot as plt import matplotlib.animation as animation #2D World -> People fluctuating around central points #Person with infection leaves infected area for dt #People on those areas get infected with probability p #People after t1 die dt = 3 space = 100 npeople = 1000 ninfected = 1 p_get_infected = .6 lifetime = 50 displacement = 3 """ Function creation movie """ def animate(i,fig,ax): ax.clear() data = np.genfromtxt('worlds/world{0:04d}'.format(i)) ax.scatter(data[:,0],data[:,1],c=data[:,2]) ax.set_xlim(0,space) ax.set_ylim(0,space) def GenerateMovie(filename,nworlds): fig = plt.figure() ax = fig.add_subplot(111) anim = animation.FuncAnimation(fig,animate,interval=40,fargs=(fig,ax),frames=nworlds) anim.save(filename,writer='ffmpeg') """ End function creation movie """ class World: #Infected area represented by 1, normal area represented by 0 def __init__(self,space): self.x = np.zeros((space,space)).astype('int') def checkAreas(self): notzeros = np.where(self.x != 0) self.x[notzeros] += 1 uninfected = np.where(self.x > 3) self.x[uninfected] = 0 def writeWorld(self,filename): output = open(filename,'w') for i in range(len(self.x)): for j in range(len(self.x)): output.write('{0:d}\t{1:d}\t{2:d}\n'.format(i,j,self.x[i,j])) output.close() class Person: def __init__(self,space): self.x = np.random.randint(space) self.y = np.random.randint(space) self.infected = 0 self.infected_time = 0 def move(self,world): self.motion() #If the area is infected the person becomes infected if world.x[self.x,self.y] != 0: if np.random.random() < p_get_infected: if self.infected == 0: self.infected = 1 #If the person is infected the area becomes infected or it's counter start over #The living time of the infected increases if self.infected != 0: world.x[self.x,self.y] = 1 self.infected_time += 1 def motion(self): dx,dy = int((np.random.randint(displacement+1))-(displacement/2)),int((np.random.randint(displacement+1))-(displacement/2)) self.x += dx self.y += dy if self.x >= space or self.x < 0: self.x = space - np.abs(self.x) if self.y >= space or self.y < 0: self.y = space - np.abs(self.y) class People: def __init__(self,npeople,world): self.people = [] for i in range(npeople): self.people.append(Person(space)) self.people[0].infected = 1 self.people[0].infected_time += 1 x,y = self.GetCoordinate(i) world.x[x,y] = 1 def MovePeople(self,world): for i in self.people: i.move(world) self.people = [i for i in self.people if i.infected_time < lifetime] def GetInfected(self): return len([i for i in self.people if i.infected>0]) def GetCoordinate(self,i): return self.people[i].x, self.people[i].y def writePeople(self,filename): output = open(filename,'w') for i in range(len(self.people)): x,y = self.GetCoordinate(i) output.write('{0:d}\t{1:d}\t{2:d}\n'.format(x,y,self.people[i].infected)) output.close() T = 10000 world = World(space) population = People(npeople,world) #Create worlds directory if it doesn't exist if not os.path.isdir('worlds'): os.mkdir('worlds') filename = 'worlds/world{0:04d}' i=0 while i < T and population.GetInfected() > 0: population.MovePeople(world) world.checkAreas() population.writePeople(filename.format(i)) i+=1 print (i,end='\r') GenerateMovie('Movie.mp4',i)
Python
2D
NSDRLIISc/e2e
e2e.py
.py
52,301
1,535
'''#!/home/magtest/test_env/bin/python''' from pymatgen.io.vasp.sets import MPRelaxSet, MPStaticSet, MPSOCSet from pymatgen import Structure from pymatgen.core.periodic_table import Element from pymatgen.analysis.magnetism.analyzer import MagneticStructureEnumerator, CollinearMagneticStructureAnalyzer from pymatgen.io.vasp.outputs import Vasprun, Chgcar, Oszicar, Outcar, Potcar from pymatgen.command_line.bader_caller import bader_analysis_from_objects import sys import os from shutil import copyfile from subprocess import Popen import datetime from time import time, sleep from ase.io import read, write from ase.build import make_supercell, sort import numpy as np from sympy import Symbol, linsolve from itertools import combinations import math from numba import jit, cuda from pickle import load, dump __author__ = "Arnab Kabiraj" __copyright__ = "Copyright 2019, NSDRL, IISc Bengaluru" __credits__ = ["Arnab Kabiraj", "Santanu Mahapatra"] root_path = os.getcwd() start_time_global = time() xc = 'PBE_54' mag_prec = 0.1 enum_prec = 0.001 max_neigh = 5 GPU_accel = False padding = True nsim = 4 kpar = 2 ncore = 1 symprec = 1e-8 d_thresh = 0.05 acc = 'default' LDAUJ_povided = {} LDAUU_povided = {} LDAUL_povided = {} with open('input') as f: for line in f: row = line.split() if 'structure_file' in line: struct_file = row[-1] elif 'XC_functional' in line: xc = row[-1] elif 'VASP_command_std' in line: cmd = line[len('VASP_command_std =')+1:-1] elif 'VASP_command_ncl' in line: cmd_ncl = line[len('VASP_command_ncl =')+1:-1] elif 'mag_prec' in line: mag_prec = float(row[-1]) elif 'enum_prec' in line: enum_prec = float(row[-1]) elif 'max_neighbors' in line: max_neigh = int(row[-1])+1 elif 'GPU_accel' in line: GPU_accel = row[-1]=='True' elif 'more_than_2_metal_layers' in line: padding = row[-1]=='True' elif 'NSIM' in line: nsim = int(row[-1]) elif 'KPAR' in line: kpar = int(row[-1]) elif 'NCORE' in line: ncore = int(row[-1]) elif 'SYMPREC' in line: symprec= float(row[-1]) elif 'LDAUJ' in line: num_spec = len(row)-2 for i in range(2,num_spec+1,2): LDAUJ_povided[row[i]] = float(row[i+1]) elif 'LDAUU' in line: num_spec = len(row)-2 for i in range(2,num_spec+1,2): LDAUU_povided[row[i]] = float(row[i+1]) elif 'LDAUL' in line: num_spec = len(row)-2 for i in range(2,num_spec+1,2): LDAUL_povided[row[i]] = float(row[i+1]) elif 'same_neighbor_thresh' in line: d_thresh = float(row[-1]) elif 'accuracy' in line: acc = row[-1] #print(LDAUU_povided) # all functions def replaceText(fileName,toFind,replaceWith): s = open(fileName).read() s = s.replace(toFind, replaceWith) f = open(fileName, 'w') f.write(s) f.close() def writeLog(string): string = str(string) f = open(root_path+'/log','a+') time = datetime.datetime.now() f.write(str(time)+' '+string+'\n') f.close() def dist_neighbors(struct): struct_l = struct.copy() struct_l.make_supercell([20,20,1]) distances = np.unique(np.sort(np.around(struct_l.distance_matrix[1],2)))[0:15] dr_max = 0.01 for i in range(len(distances)): for j in range(len(distances)): dr = np.abs(distances[i]-distances[j]) if distances[j]<distances[i] and dr<d_thresh: distances[i]=distances[j] if dr>dr_max: dr_max = dr distances = np.unique(distances) msg = 'neighbor distances are: '+str(distances)+' ang' print(msg) writeLog(msg) msg = 'treating '+str(dr_max)+' ang separated atoms as same neighbors' print(msg) writeLog(msg) distances[0]=dr_max return distances def Nfinder(struct_mag,site,d_N,dr): N = len(struct_mag) coord_site = struct_mag.cart_coords[site] Ns = struct_mag.get_neighbors_in_shell(coord_site,d_N,dr) Ns_wrapped = Ns[:] candidates = Ns[:] for i in range(len(Ns)): Ns_wrapped[i] = Ns[i][0].to_unit_cell() for j in range(N): if struct_mag[j].distance(Ns_wrapped[i])<0.01: candidates[i] = j break return candidates @cuda.jit def my_kernel(all_coords,coord_N,index): """ Code for kernel. """ pos = cuda.grid(1) if pos < all_coords.size: if math.sqrt((all_coords[pos]-coord_N[0])**2 + (all_coords[pos+1]-coord_N[1])**2 + (all_coords[pos+2]-coord_N[2])**2) < 0.01: index[0] = pos/3 def NfinderGPU(struc_mag,site, d_N, dr): coord_site = struc_mag.cart_coords[site] Ns = struc_mag.get_neighbors_in_shell(coord_site,d_N,dr) #print(Ns) Ns_wrapped = Ns[:] candidates = Ns[:] for i in range(len(Ns)): Ns_wrapped[i] = Ns[i][0].to_unit_cell() coord_N = np.array([Ns_wrapped[i].x,Ns_wrapped[i].y,Ns_wrapped[i].z],dtype='float32') index = np.array([-5]) threadsperblock = 480 blockspergrid = math.ceil(all_coords.shape[0] / threadsperblock) my_kernel[blockspergrid,threadsperblock](all_coords,coord_N,index) candidates[i]=index[0] return candidates def find_max_len(lst): maxList = max(lst, key = lambda i: len(i)) maxLength = len(maxList) return maxLength def make_homogenous(lst): msg = 'finding and padding neighbors' print(msg) writeLog(msg) max_len = find_max_len(lst) for i in range(len(lst)): if len(lst[i])<max_len: pad = [100000]*(max_len-len(lst[i])) lst[i] += pad print(str(i)+'p / '+str(len(lst)-1)) @jit(nopython=True) def tFunc(spin_abs,spin_x,spin_y,spin_z,mags,magsqs,T,J2flag,J3flag,J4flag,J5flag): for t in range(trange): mag = 0 for i in range(N): site = np.random.randint(0,N) N1s = N1list[site] N2s = N2list[site] N3s = N3list[site] N4s = N4list[site] N5s = N5list[site] S_current = np.array([spin_x[site],spin_y[site],spin_z[site]]) u, v = np.random.random(),np.random.random() phi = 2*np.pi*u theta = np.arccos(2*v-1) S_x = spin_abs[site]*np.sin(theta)*np.cos(phi) S_y = spin_abs[site]*np.sin(theta)*np.sin(phi) S_z = spin_abs[site]*np.cos(theta) S_after = np.array([S_x,S_y,S_z]) E_current = 0 E_after = 0 for N1 in N1s: if N1!=100000 or N1!=-5: S_N1 = np.array([spin_x[N1],spin_y[N1],spin_z[N1]]) E_current += -J1*np.dot(S_current,S_N1) E_after += -J1*np.dot(S_after,S_N1) if J2flag: for N2 in N2s: if N2!=100000 or N2!=-5: S_N2 = np.array([spin_x[N2],spin_y[N2],spin_z[N2]]) E_current += -J2*np.dot(S_current,S_N2) E_after += -J2*np.dot(S_after,S_N2) if J3flag: for N3 in N3s: if N3!= 100000 or N3!=-5: S_N3 = np.array([spin_x[N3],spin_y[N3],spin_z[N3]]) E_current += -J3*np.dot(S_current,S_N3) E_after += -J3*np.dot(S_after,S_N3) if J4flag: for N4 in N4s: if N4!= 100000 or N4!=-5: S_N4 = np.array([spin_x[N4],spin_y[N4],spin_z[N4]]) E_current += -J4*np.dot(S_current,S_N4) E_after += -J4*np.dot(S_after,S_N4) if J5flag: for N5 in N5s: if N5!= 100000 or N5!=-5: S_N5 = np.array([spin_x[N5],spin_y[N5],spin_z[N5]]) E_current += -J5*np.dot(S_current,S_N5) E_after += -J5*np.dot(S_after,S_N5) E_current += k_x*np.square(S_current[0]) + k_y*np.square(S_current[1]) + k_z*np.square(S_current[2]) E_after += k_x*np.square(S_x) + k_y*np.square(S_y) + k_z*np.square(S_z) del_E = E_after-E_current if del_E < 0: spin_x[site],spin_y[site],spin_z[site] = S_x,S_y,S_z else: samp = np.random.random() if samp <= np.exp(-del_E/(kB*T)): spin_x[site],spin_y[site],spin_z[site] = S_x,S_y,S_z if t>=threshold: mag_vec = 2*np.array([np.sum(spin_x),np.sum(spin_y),np.sum(spin_z)]) mag = np.linalg.norm(mag_vec) mags[t-threshold]=np.abs(mag) magsqs[t-threshold]=np.square(mag) return np.mean(mags),np.mean(magsqs) # main code msg = '*'*150 print(msg) writeLog(msg) msg = '*** this code have been developed by Arnab Kabiraj at Nano-Scale Device Research Laboratory (NSDRL), IISc, Bengaluru, India ***\n' msg += '*** for any queries please contact the authors at kabiraj@iisc.ac.in or santanu@iisc.ac.in ***' print(msg) writeLog(msg) msg = '*'*150 print(msg) writeLog(msg) if acc == 'high': msg = '* command for high accuracy detected, the calculations could take significantly more time than ususal\n' msg += '* make sure the number of cores is an integer multiple of 4' print(msg) writeLog(msg) cell = read(struct_file) c = cell.get_cell_lengths_and_angles()[2] for i in range(len(cell)): if cell[i].z > c*0.75: cell[i].z = cell[i].z - c cell.center(12.75,2) write('2D.xsf', sort(cell)) struct = Structure.from_file('2D.xsf') mag_enum = MagneticStructureEnumerator(struct,transformation_kwargs={'symm_prec':mag_prec,'enum_precision_parameter':enum_prec}) mag_structs = [] for s_mag in mag_enum.ordered_structures: if (s_mag.lattice.c < struct.lattice.c+3.0 and s_mag.lattice.c > struct.lattice.c-3.0): mag_structs.append(s_mag) num_struct = len(mag_structs) if num_struct == 1: msg = '*** only one config could be generated, exiting, try out a new material' print(msg) writeLog(msg) sys.exit() msg = 'total '+str(num_struct)+' configs generated' print(msg) writeLog(msg) sleep_time = 45 num_atoms = [] for struct in mag_structs: num_atoms.append(len(struct)) lcm_atoms = np.lcm.reduce(num_atoms) num_valid_configs = 0 magnetic_list = [Element('Co'), Element('Cr'), Element('Fe'), Element('Mn'), Element('Mo'), Element('Ni'), Element('V'), Element('W'), Element('Ce'), Element('Os'), Element('Sc'), Element('Ti'), Element('Ag'), Element('Zr'), Element('Pd'), Element('Rh'), Element('Hf'), Element('Nb'), Element('Y'), Element('Re'), Element('Cu'), Element('Ru'), Element('Pt'), Element('La')] LDAUJ_dict = {'Co': 0, 'Cr': 0, 'Fe': 0, 'Mn': 0, 'Mo': 0, 'Ni': 0, 'V': 0, 'W': 0, 'Nb': 0, 'Sc': 0, 'Ru': 0, 'Rh': 0, 'Pd': 0, 'Cu': 0, 'Y': 0, 'Os': 0, 'Ti': 0, 'Zr': 0, 'Re': 0, 'Hf': 0, 'Pt':0, 'La':0} if LDAUJ_povided: LDAUJ_dict.update(LDAUJ_povided) LDAUU_dict = {'Co': 3.32, 'Cr': 3.7, 'Fe': 5.3, 'Mn': 3.9, 'Mo': 4.38, 'Ni': 6.2, 'V': 3.25, 'W': 6.2, 'Nb': 1.45, 'Sc': 4.18, 'Ru': 4.29, 'Rh': 4.17, 'Pd': 2.96, 'Cu': 7.71, 'Y': 3.23, 'Os': 2.47, 'Ti': 5.89, 'Zr': 5.55, 'Re': 1.28, 'Hf': 4.77, 'Pt': 2.95, 'La':5.3} if LDAUU_povided: LDAUU_dict.update(LDAUU_povided) LDAUL_dict = {'Co': 2, 'Cr': 2, 'Fe': 2, 'Mn': 2, 'Mo': 2, 'Ni': 2, 'V': 2, 'W': 2, 'Nb': 2, 'Sc': 2, 'Ru': 2, 'Rh': 2, 'Pd': 2, 'Cu': 2, 'Y': 2, 'Os': 2, 'Ti': 2, 'Zr': 2, 'Re': 2, 'Hf': 2, 'Pt':2, 'La':2} if LDAUL_povided: LDAUL_dict.update(LDAUL_povided) relx_dict = {'ISMEAR': 0, 'SIGMA': 0.01, 'ISIF': 4, 'EDIFF': 1E-4, 'POTIM': 0.3, 'EDIFFG': -0.01, 'SYMPREC': 1E-8, 'KPAR': kpar, 'NCORE': ncore, 'NSIM': nsim, 'LCHARG': False, 'ICHARG': 2, 'LDAU': True, 'LDAUJ': LDAUJ_dict, 'LDAUL': LDAUL_dict, 'LDAUU': LDAUU_dict, 'LWAVE': False, 'LDAUPRINT': 1, 'LDAUTYPE': 2, 'LASPH': True, 'LMAXMIX': 4} stat_dict = {'ISMEAR': -5, 'EDIFF': 1E-6, 'SYMPREC': 1E-8, 'KPAR': kpar, 'NCORE': ncore, 'NSIM': nsim, 'ICHARG': 2, 'LDAU': True, 'LDAUJ': LDAUJ_dict, 'LDAUL': LDAUL_dict, 'LDAUU': LDAUU_dict, 'NELM': 120, 'LVHAR': False, 'LDAUPRINT': 1, 'LDAUTYPE': 2, 'LASPH': True, 'LMAXMIX': 4, 'LCHARG': True, 'LWAVE': False, 'LVTOT': False} mae_dict = {'ISMEAR': -5, 'EDIFF': 1E-8, 'SYMPREC': 1E-8, 'KPAR': kpar, 'NCORE': ncore, 'NSIM': nsim, 'LORBMOM': True, 'LAECHG': False, 'LDAU': True, 'LDAUJ': LDAUJ_dict, 'LDAUL': LDAUL_dict, 'LDAUU': LDAUU_dict,'NELMIN': 6, 'ICHARG': 1, 'NELM': 250, 'LVHAR': False, 'LDAUPRINT': 1, 'LDAUTYPE': 2, 'LASPH': True, 'LMAXMIX': 4, 'LCHARG': True, 'LWAVE': True, 'ISYM': -1, 'LVTOT': False} if acc == 'high': relx_dict['ALGO'] = 'Normal' relx_dict['PREC'] = 'Accurate' relx_dict['EDIFF'] = 1E-5 relx_dict['KPAR'] = 4 stat_dict['KPAR'] = 4 mae_dict['KPAR'] = 4 start_time_dft = time() for i in range(num_struct): struct_current = mag_structs[i].get_sorted_structure() mag_tot = 0 for j in range(len(struct_current)): element = struct_current[j].specie.element if element in magnetic_list: try: mag_tot += struct_current[j].specie.spin except Exception: msg = '** the structure '+str(i)+' has uneven spins, continuing without it' print(msg) writeLog(msg) mag_tot = 1000 break if i>0 and np.abs(mag_tot) > 0.1: msg = '** the structure '+str(i)+' seems ferrimagnetic, continuing without it' print(msg) writeLog(msg) continue submission = 0 p = 0 relx_path = root_path+'/config_'+str(i)+'/relx' n = len(struct_current) spins = [0]*n for j in range(n): try: spins[j] = struct_current.species[j].spin except Exception: spins[j] = 0.0 factor = float(lcm_atoms)/n if acc == 'high': relx = MPRelaxSet(struct_current,user_incar_settings=relx_dict,user_kpoints_settings={'reciprocal_density':300},force_gamma=True,user_potcar_functional=xc) else: relx = MPRelaxSet(struct_current,user_incar_settings=relx_dict,force_gamma=True,user_potcar_functional=xc) relx.write_input(relx_path) num_valid_configs += 1 while 1: if submission>2: nsim_old = nsim nsim = max([4,nsim-4]) replaceText(relx_path+'/INCAR','NSIM = '+str(nsim_old),'NSIM = '+str(nsim)) msg = 'current NSIM is '+str(nsim_old)+', it will be changed to '+str(nsim)+' in this run' print(msg) writeLog(msg) if submission==5: replaceText(relx_path+'/INCAR','POTIM = 0.3','POTIM = 0.1') msg = 'reducing POTIM to 0.1' print(msg) writeLog(msg) elif submission==7: replaceText(relx_path+'/INCAR','POTIM = 0.1','POTIM = 0.05') msg = 'reducing POTIM to 0.05' print(msg) writeLog(msg) elif submission==9: replaceText(relx_path+'/INCAR','POTIM = 0.05','POTIM = 0.01') msg = 'reducing POTIM to 0.01' print(msg) writeLog(msg) elif submission==11: replaceText(relx_path+'/INCAR','SYMPREC = 1e-08','SYMPREC = 1e-05') msg = 'increasing SYMPREC' print(msg) writeLog(msg) elif submission==14: f = open(relx_path+'/INCAR','a+') f.write('\nISYM = 0') f.close() msg = 'lots of calculation have failed, turning off symmetry' print(msg) writeLog(msg) try: msg = 'checking vasp run status for config_'+str(i)+' relaxation' print(msg) writeLog(msg) try: temp_struct = Structure.from_file(relx_path+'/CONTCAR') copyfile(relx_path+'/CONTCAR',relx_path+'/CONTCAR.last') copyfile(relx_path+'/CONTCAR',relx_path+'/POSCAR') msg = 'copied CONTCAR to POSCAR' print(msg) writeLog(msg) except Exception: msg = 'no CONTCAR so far' print(msg) writeLog(msg) run = Vasprun(relx_path+'/vasprun.xml',parse_dos=False,parse_eigen=False) if run.converged: msg = 'relaxation finished' print(msg) writeLog(msg) break else: msg = 'relaxation have not converged, resubmitting' print(msg) writeLog(msg) raise ValueError except Exception as e: print(e) writeLog(e) os.chdir(relx_path) p = (Popen(cmd, shell=True)) os.chdir(root_path) submission += 1 msg = 'submitted relaxation for config_'+str(i)+'' print(msg) writeLog(msg) msg = 'this is submission #'+str(submission)+' for this config' print(msg) writeLog(msg) while p.poll() is None: msg = 'job running, waiting..' print(msg) sleep(sleep_time) path = root_path+'/config_'+str(i)+'/stat' stat_struct = Structure.from_file(relx_path+'/CONTCAR',sort=True) stat_struct.add_spin_by_site(spins) if acc == 'high': stat = MPStaticSet(stat_struct,user_incar_settings=stat_dict,reciprocal_density=1000,force_gamma=True,user_potcar_functional=xc) else: stat = MPStaticSet(stat_struct,user_incar_settings=stat_dict,reciprocal_density=300,force_gamma=True,user_potcar_functional=xc) stat.write_input(path) p = 0 submission = 0 while 1: if submission>2: nsim_old = nsim nsim = max([4,nsim-4]) replaceText(path+'/INCAR','NSIM = '+str(nsim_old),'NSIM = '+str(nsim)) msg = 'current NSIM is '+str(nsim_old)+', it will be changed to '+str(nsim)+' in this run' print(msg) writeLog(msg) if submission==5: f = open(path+'/INCAR','a+') f.write('\nAMIX = 0.2\nBMIX = 0.0001\nAMIX_MAG = 0.8\nBMIX_MAG = 0.0001') f.close() msg = 'lots of calculation have failed, switching to Linear Mixing' print(msg) writeLog(msg) os.remove(path+'/WAVECAR') os.remove(path+'/CHGCAR') elif submission==7: replaceText(path+'/INCAR','SYMPREC = 1e-08','SYMPREC = 1e-05') msg = 'increasing SYMPREC' print(msg) writeLog(msg) elif submission==9: f = open(path+'/INCAR','a+') f.write('\nISYM = 0') f.close() msg = 'lots of calculation have failed, turning off symmetry' print(msg) writeLog(msg) try: msg = 'checking vasp run status for config_'+str(i)+' static run' print(msg) writeLog(msg) run = Vasprun(path+'/vasprun.xml',parse_dos=False,parse_eigen=False) if run.converged_electronic: msg = 'static run finished' print(msg) writeLog(msg) energy = float(run.final_energy) energy = energy*factor if i==0: E_FM = energy elif energy<E_FM: msg = 'config_'+str(i)+' is more stable than FM' print(msg) writeLog(msg) msg = '*** this material is NOT FM, exiting, try out a new material' print(msg) writeLog(msg) sys.exit() osz = Oszicar(path+'/OSZICAR') config_mag = float(osz.ionic_steps[-1]['mag']) if i==0 and np.abs(config_mag)<1.0: msg = '** too low magnetization ('+str(config_mag)+') for an FM config_'+str(i)+', check OUTCAR' print(msg) writeLog(msg) elif i>0 and np.abs(config_mag)>0.05: msg = '** too large magnetization ('+str(config_mag)+') for an AFM config_'+str(i)+', check OUTCAR' print(msg) writeLog(msg) try: charge = Chgcar.from_file(path+'/CHGCAR') except Exception: charge = Chgcar.from_file(path+'/CHGCAR.sp') if charge.is_spin_polarized: try: os.rename(path+'/CHGCAR',path+'/CHGCAR.sp') msg = 'renamed CHGCAR to CHGCAR.sp' print(msg) writeLog(msg) except Exception: msg = 'CHGCAR.sp found' else: msg = '*** no spin-polarized CHGCAR found for config_'+str(i)+' static run' print(msg) writeLog(msg) break else: msg = 'static run have not converged, resubmitting..' print(msg) writeLog(msg) raise ValueError except Exception as e: print(e) writeLog(e) os.chdir(path) p = (Popen(cmd, shell=True)) os.chdir(root_path) submission += 1 msg = 'submitted static_run for config_'+str(i)+'' print(msg) writeLog(msg) msg = 'this is submission #'+str(submission)+' for this config' print(msg) writeLog(msg) while p.poll() is None: msg = 'job running, waiting..' print(msg) sleep(sleep_time) end_time_dft = time() time_dft = np.around(end_time_dft - start_time_dft, 2) msg = 'all relaxations and static runs have finished gracefully, proceeding to MAE calculations now' print(msg) writeLog(msg) msg = 'DFT energy calculations of all possible configurations took total '+str(time_dft)+' s' print(msg) writeLog(msg) submission = 0 p = 0 pre_path = root_path+'/config_0/stat' path_coll = root_path+'/MAE/coll' if acc == 'high': stat = MPStaticSet.from_prev_calc(pre_path,user_incar_settings=mae_dict,reciprocal_density=1000,force_gamma=True,user_potcar_functional=xc) else: stat = MPStaticSet.from_prev_calc(pre_path,user_incar_settings=mae_dict,reciprocal_density=300,force_gamma=True,user_potcar_functional=xc) stat.write_input(path_coll) if os.path.isfile(path_coll+'/WAVECAR'): msg = 'collinear WAVECAR exists' print(msg) writeLog(msg) if not os.path.isfile(path_coll+'/CHGCAR'): try: copyfile(pre_path+'/CHGCAR.sp',path_coll+'/CHGCAR') except: msg = 'no CHGCAR found in stat, continuing but it might fail' print(msg) writeLog(msg) else: msg = 'collinear CHGCAR exists' print(msg) writeLog(msg) start_time_mae = time() while 1: if submission>2: nsim_old = nsim nsim = max([4,nsim-4]) replaceText(path_coll+'/INCAR','NSIM = '+str(nsim_old),'NSIM = '+str(nsim)) msg = 'current NSIM is '+str(nsim_old)+', it will be changed to '+str(nsim)+' in this run' print(msg) writeLog(msg) if submission==5: f = open(path_coll+'/INCAR','a+') f.write('\nAMIX = 0.2\nBMIX = 0.0001\nAMIX_MAG = 0.8\nBMIX_MAG = 0.0001') f.close() msg = 'lots of calculation have failed, switching to Linear Mixing' print(msg) writeLog(msg) os.remove(path_coll+'/WAVECAR') os.remove(path_coll+'/CHGCAR') try: copyfile(pre_path+'/CHGCAR',path_coll+'/CHGCAR') except: msg = 'no CHGCAR found in stat, continuing but it might fail' print(msg) writeLog(msg) try: msg = 'checking vasp run status for collinear' print(msg) writeLog(msg) run = Vasprun(path_coll+'/vasprun.xml',parse_dos=False,parse_eigen=False) if run.converged_electronic: msg = 'collinear run finished' print(msg) writeLog(msg) break else: msg = 'collinear run have not converged, resubmitting..' print(msg) writeLog(msg) raise ValueError except: os.chdir(path_coll) p = (Popen(cmd, shell=True)) os.chdir(root_path) submission += 1 msg = 'submitted collinear run' print(msg) writeLog(msg) msg = 'this is submission #'+str(submission) print(msg) writeLog(msg) while p.poll() is None: msg = 'job running, waiting..' print(msg) sleep(sleep_time) saxes = [(1,0,0),(0,1,0),(1,1,0),(0,0,1)] for axis in saxes: path = root_path+'/MAE/'+str(axis).replace(' ','') submission = 0 p = 0 if acc == 'high': soc = MPSOCSet.from_prev_calc(path_coll,saxis=axis,nbands_factor=2,reciprocal_density=1000,force_gamma=True,user_potcar_functional=xc) else: soc = MPSOCSet.from_prev_calc(path_coll,saxis=axis,nbands_factor=2,reciprocal_density=300,force_gamma=True,user_potcar_functional=xc) soc.write_input(path) replaceText(path+'/INCAR','LCHARG = True','LCHARG = False') replaceText(path+'/INCAR','LWAVE = True','LWAVE = False') if acc == 'high': replaceText(path+'/INCAR','LVTOT = False','KPAR = 4') else: replaceText(path+'/INCAR','LVTOT = False','KPAR = 2') try: copyfile(path_coll+'/WAVECAR',path+'/WAVECAR') except: msg = '*** no collinear WAVECAR found, exiting' print(msg) writeLog(msg) raise ValueError while 1: if submission>2: nsim_old = nsim nsim = max([4,nsim-4]) replaceText(path+'/INCAR','NSIM = '+str(nsim_old),'NSIM = '+str(nsim)) msg = 'current NSIM is '+str(nsim_old)+', it will be changed to '+str(nsim)+' in this run' print(msg) writeLog(msg) if submission==7: f = open(path+'/INCAR','a+') f.write('\nAMIX = 0.2\nBMIX = 0.0001\nAMIX_MAG = 0.8\nBMIX_MAG = 0.0001') f.close() msg = 'lots of calculation have failed, applying Linear Mixing' print(msg) writeLog(msg) try: msg = 'checking vasp run status for non-collinear '+str(axis) print(msg) writeLog(msg) run = Vasprun(path+'/vasprun.xml',parse_dos=False,parse_eigen=False) if run.converged_electronic: msg = 'non-collinear run finished for '+str(axis) print(msg) writeLog(msg) try: os.remove(path+'/CHGCAR') msg = 'removed CHGCAR' print(msg) writeLog(msg) except: msg = 'no CHGCAR found to remove' print(msg) writeLog(msg) try: os.remove(path+'/WAVECAR') msg = 'removed WAVECAR' print(msg) writeLog(msg) except: msg = 'no WAVERCAR found to remove' print(msg) writeLog(msg) break else: msg = 'non-collinear run have not converged for '+str(axis)+', resubmitting..' print(msg) writeLog(msg) raise ValueError except: os.chdir(path) p = (Popen(cmd_ncl, shell=True)) os.chdir(root_path) submission += 1 msg = 'submitted non-collinear run for '+str(axis) print(msg) writeLog(msg) msg = 'this is submission #'+str(submission)+' for this axis' print(msg) writeLog(msg) while p.poll() is None: msg = 'job running, waiting..' print(msg) sleep(sleep_time) end_time_mae = time() time_mae = np.around(end_time_mae - start_time_mae, 2) msg='all MAE calculations finished, attempting to fit the Hamiltonian now' print(msg) writeLog(msg) msg = 'the MAE calculations took '+str(time_mae)+' s' print(msg) writeLog(msg) E0 = Symbol('E0') J1 = Symbol('J1') J2 = Symbol('J2') J3 = Symbol('J3') J4 = Symbol('J4') J5 = Symbol('J5') kB = np.double(8.6173303e-5) num_neigh = min([max_neigh, num_valid_configs]) msg = 'total '+str(num_valid_configs)+' valid FM/AFM configs have been detected, including '+str(num_neigh)+' nearest-neighbors in the fitting' print(msg) writeLog(msg) fitted = False semifinal_list = [] for i in range(num_struct): path = root_path+'/config_'+str(i)+'/stat' msg = 'checking vasp run status for config_'+str(i)+' static run' print(msg) writeLog(msg) struct = mag_structs[i].get_sorted_structure() #print(struct) mag_tot = 0 for j in range(len(struct)): element = struct[j].specie.element if element in magnetic_list: try: mag_tot += struct[j].specie.spin except Exception: msg = '** the structure '+str(i)+' has uneven spins, continuing without it' print(msg) writeLog(msg) mag_tot = 1000 break if i>0 and np.abs(mag_tot) > 0.1: msg = '** the structure '+str(i)+' seems ferrimagnetic, continuing without it' print(msg) writeLog(msg) continue run = Vasprun(path+'/vasprun.xml',parse_dos=False,parse_eigen=False) if not run.converged_electronic: msg = '*** static run have not converged for config_'+str(i)+', exiting' print(msg) writeLog(msg) raise ValueError else: msg = 'found converged static run' print(msg) writeLog(msg) energy = float(run.final_energy) factor = float(lcm_atoms)/len(struct) if factor!=int(factor): msg = '*** factor is float, '+str(factor)+', exiting' print(msg) writeLog(msg) raise ValueError energy = energy*factor if i==0: E_FM = energy elif energy<E_FM: msg = 'config_'+str(i)+' is more stable than FM' print(msg) writeLog(msg) msg = '*** this material is NOT FM' print(msg) writeLog(msg) semifinal_list.append((path,energy,struct)) semifinal_list = sorted(semifinal_list, key = lambda x : x[1]) while num_neigh>=2: if len(semifinal_list)>num_neigh: final_list = semifinal_list[0:num_neigh] else: final_list = semifinal_list[:] num_struct = len(final_list) eqn_set = [0]*num_struct for i in range(num_struct): path = final_list[i][0] energy = final_list[i][1] struct = final_list[i][2] factor = float(lcm_atoms)/len(struct) config = path[-13:-5] if '_0' in config: if not os.path.exists(path+'/bader.dat'): chgcar = Chgcar.from_file(path+'/CHGCAR.sp') if not chgcar.is_spin_polarized: msg = '** '+path+'/CHGCAR.sp is not spin-polarized, exiting' print(msg) writeLog(msg) raise ValueError potcar = Potcar.from_file(path+'/POTCAR') aeccar0 = Chgcar.from_file(path+'/AECCAR0') aeccar2 = Chgcar.from_file(path+'/AECCAR2') msg = 'starting bader analysis for '+config print(msg) writeLog(msg) ba = bader_analysis_from_objects(chgcar=chgcar,potcar=potcar,aeccar0=aeccar0,aeccar2=aeccar2) msg = 'finished bader analysis successfully' print(msg) writeLog(msg) f = open(path+'/bader.dat','wb') dump(ba,f) f.close() magmom_FM = max(ba['magmom']) else: f = open(path+'/bader.dat','rb') ba = load(f) f.close() magmom_FM = max(ba['magmom']) msg = 'read magmoms from file' print(msg) writeLog(msg) osz = Oszicar(path+'/OSZICAR') out = Outcar(path+'/OUTCAR') config_mag = float(osz.ionic_steps[-1]['mag']) if i==0 and np.abs(config_mag)<1.0: msg = '** too low magnetization ('+str(config_mag)+') for an FM config_'+str(i)+', check OUTCAR' print(msg) writeLog(msg) elif i>0 and np.abs(config_mag)>0.05: msg = '** too large magnetization ('+str(config_mag)+') for an AFM config_'+str(i)+', check OUTCAR' print(msg) writeLog(msg) sites_mag = [] magmoms_mag = [] magmoms_out = [] for j in range(len(struct)): element = struct[j].specie.element if element in magnetic_list: sign_magmom = np.sign(struct[j].specie.spin) magmom = sign_magmom*magmom_FM magmoms_mag.append(magmom) sites_mag.append(struct[j]) magmoms_out.append(out.magnetization[j]['tot']) struct_mag = Structure.from_sites(sites_mag) struct_mag_out = Structure.from_sites(sites_mag) struct_mag.remove_spin() struct_mag.add_site_property('magmom',magmoms_mag) struct_mag_out.add_site_property('magmom',magmoms_out) N = len(struct_mag) msg = config+' with scaling factor '+str(factor)+' = ' print(msg) writeLog(msg) print(struct_mag) writeLog(struct_mag) msg = 'same config with magmoms from OUTCAR is printed below, make sure this does not deviate too much from above' print(msg) writeLog(msg) print(struct_mag_out) writeLog(struct_mag_out) if i==0: S_FM = magmom_FM/2.0 N_FM = float(len(struct_mag)) ds = dist_neighbors(struct_mag) dr = ds[0] eqn = E0 - energy for j in range(N): site = j S_site = struct_mag.site_properties['magmom'][j]/2.0 if num_struct==2: N1s = Nfinder(struct_mag,site,ds[1],dr) N2s = [] N3s = [] N4s = [] N5s = [] elif num_struct==3: N1s = Nfinder(struct_mag,site,ds[1],dr) N2s = Nfinder(struct_mag,site,ds[2],dr) N3s = [] N4s = [] N5s = [] elif num_struct==4: N1s = Nfinder(struct_mag,site,ds[1],dr) N2s = Nfinder(struct_mag,site,ds[2],dr) N3s = Nfinder(struct_mag,site,ds[3],dr) N4s = [] N5s = [] elif num_struct==5: N1s = Nfinder(struct_mag,site,ds[1],dr) N2s = Nfinder(struct_mag,site,ds[2],dr) N3s = Nfinder(struct_mag,site,ds[3],dr) N4s = Nfinder(struct_mag,site,ds[4],dr) N5s = [] elif num_struct==6: N1s = Nfinder(struct_mag,site,ds[1],dr) N2s = Nfinder(struct_mag,site,ds[2],dr) N3s = Nfinder(struct_mag,site,ds[3],dr) N4s = Nfinder(struct_mag,site,ds[4],dr) N5s = Nfinder(struct_mag,site,ds[5],dr) for N1 in N1s: S_N1 = struct_mag.site_properties['magmom'][N1]/2.0 eqn += -0.5*J1*S_site*S_N1*factor if N2s: for N2 in N2s: S_N2 = struct_mag.site_properties['magmom'][N2]/2.0 eqn += -0.5*J2*S_site*S_N2*factor if N3s: for N3 in N3s: S_N3 = struct_mag.site_properties['magmom'][N3]/2.0 eqn += -0.5*J3*S_site*S_N3*factor if N4s: for N4 in N4s: S_N4 = struct_mag.site_properties['magmom'][N4]/2.0 eqn += -0.5*J4*S_site*S_N4*factor if N5s: for N5 in N5s: S_N5 = struct_mag.site_properties['magmom'][N5]/2.0 eqn += -0.5*J5*S_site*S_N5*factor eqn_set[i] = eqn msg = 'mu = '+str(magmom_FM)+' bohr magnetron/magnetic atom' print(msg) writeLog(msg) msg = 'eqns are:' print(msg) writeLog(msg) for eqn in eqn_set: msg = str(eqn)+' = 0' print(msg) writeLog(msg) if num_struct == 2: soln = linsolve(eqn_set, E0, J1) elif num_struct == 3: soln = linsolve(eqn_set, E0, J1, J2) elif num_struct == 4: soln = linsolve(eqn_set, E0, J1, J2, J3) elif num_struct == 5: soln = linsolve(eqn_set, E0, J1, J2, J3, J4) elif num_struct == 6: soln = linsolve(eqn_set, E0, J1, J2, J3, J4, J5) soln = list(soln) msg = 'the solutions are:' print(msg) writeLog(msg) print(soln) writeLog(soln) if soln and np.max(np.abs(soln[0]))<1e3: fitted = True break else: num_neigh -= 1 msg = 'looks like these set of equations are either not solvable or yielding unphysical values' print(msg) writeLog(msg) msg = 'reducing the number of included NNs to '+str(num_neigh) print(msg) writeLog(msg) if not fitted: msg = '*** could not fit the Hamiltonian, exiting' print(msg) writeLog(msg) sys.exit() if num_struct == 2: E0 = soln[0][0] J1 = soln[0][1] J2 = 0 J3 = 0 J4 = 0 J5 = 0 msg = 'the solutions are:' print(msg) writeLog(msg) msg = 'E0 = '+str(E0)+' eV' print(msg) writeLog(msg) msg = 'J1 = '+str(J1*1e3)+' meV/link with d1 = '+str(ds[1])+' ang and NN coordination = '+str(len(N1s)) print(msg) writeLog(msg) elif num_struct == 3: E0 = soln[0][0] J1 = soln[0][1] J2 = soln[0][2] J3 = 0 J4 = 0 J5 = 0 msg = 'the solutions are:' print(msg) writeLog(msg) msg = 'E0 = '+str(E0)+' eV' print(msg) writeLog(msg) msg = 'J1 = '+str(J1*1e3)+' meV/link with d1 = '+str(ds[1])+' ang and NN coordination = '+str(len(N1s)) print(msg) writeLog(msg) msg = 'J2 = '+str(J2*1e3)+' meV/link with d2 = '+str(ds[2])+' ang and NNN coordination = '+str(len(N2s)) print(msg) writeLog(msg) elif num_struct == 4: E0 = soln[0][0] J1 = soln[0][1] J2 = soln[0][2] J3 = soln[0][3] J4 = 0 J5 = 0 msg = 'the solutions are:' print(msg) writeLog(msg) msg = 'E0 = '+str(E0)+' eV' print(msg) writeLog(msg) msg = 'J1 = '+str(J1*1e3)+' meV/link with d1 = '+str(ds[1])+' ang and NN coordination = '+str(len(N1s)) print(msg) writeLog(msg) msg = 'J2 = '+str(J2*1e3)+' meV/link with d2 = '+str(ds[2])+' ang and NNN coordination = '+str(len(N2s)) print(msg) writeLog(msg) msg = 'J3 = '+str(J3*1e3)+' meV/link with d3 = '+str(ds[3])+' ang and NNNN coordination = '+str(len(N3s)) print(msg) writeLog(msg) elif num_struct == 5: E0 = soln[0][0] J1 = soln[0][1] J2 = soln[0][2] J3 = soln[0][3] J4 = soln[0][4] J5 = 0 msg = 'the solutions are:' print(msg) writeLog(msg) msg = 'E0 = '+str(E0)+' eV' print(msg) writeLog(msg) msg = 'J1 = '+str(J1*1e3)+' meV/link with d1 = '+str(ds[1])+' ang and NN coordination = '+str(len(N1s)) print(msg) writeLog(msg) msg = 'J2 = '+str(J2*1e3)+' meV/link with d2 = '+str(ds[2])+' ang and NNN coordination = '+str(len(N2s)) print(msg) writeLog(msg) msg = 'J3 = '+str(J3*1e3)+' meV/link with d3 = '+str(ds[3])+' ang and NNNN coordination = '+str(len(N3s)) print(msg) writeLog(msg) msg = 'J4 = '+str(J4*1e3)+' meV/link with d4 = '+str(ds[4])+' ang and NNNNN coordination = '+str(len(N4s)) print(msg) writeLog(msg) elif num_struct == 6: E0 = soln[0][0] J1 = soln[0][1] J2 = soln[0][2] J3 = soln[0][3] J4 = soln[0][4] J5 = soln[0][5] msg = 'the solutions are:' print(msg) writeLog(msg) msg = 'E0 = '+str(E0)+' eV' print(msg) writeLog(msg) msg = 'J1 = '+str(J1*1e3)+' meV/link with d1 = '+str(ds[1])+' ang and NN coordination = '+str(len(N1s)) print(msg) writeLog(msg) msg = 'J2 = '+str(J2*1e3)+' meV/link with d2 = '+str(ds[2])+' ang and NNN coordination = '+str(len(N2s)) print(msg) writeLog(msg) msg = 'J3 = '+str(J3*1e3)+' meV/link with d3 = '+str(ds[3])+' ang and NNNN coordination = '+str(len(N3s)) print(msg) writeLog(msg) msg = 'J4 = '+str(J4*1e3)+' meV/link with d4 = '+str(ds[4])+' ang and NNNNN coordination = '+str(len(N4s)) print(msg) writeLog(msg) msg = 'J5 = '+str(J5*1e3)+' meV/link with d5 = '+str(ds[5])+' ang and NNNNNN coordination = '+str(len(N5s)) print(msg) writeLog(msg) if ds[1]/ds[2] >= 0.8: msg = '** d1/d2 is greater than 0.8, consider adding the 2nd neighbor for accurate results' print(msg) writeLog(msg) elif ds[1]/ds[3] >= 0.7: msg = '** d1/d3 is greater than 0.7, consider adding the 3rd neighbor for accurate results' print(msg) writeLog(msg) saxes = [(1,0,0),(0,1,0),(1,1,0),(0,0,1)] energies_ncl = [0]*len(saxes) for i in range(len(saxes)): axis = saxes[i] path = root_path+'/MAE/'+str(axis).replace(' ','') msg = 'checking vasp run status for non-collinear '+str(axis) print(msg) writeLog(msg) run = Vasprun(path+'/vasprun.xml',parse_dos=False,parse_eigen=False) energies_ncl[i] = float(run.final_energy) EMA = saxes[np.argmin(energies_ncl)] msg = 'the easy magnetization axis is '+str(EMA) print(msg) writeLog(msg) E_100_001 = (energies_ncl[0] - energies_ncl[3])/N_FM E_010_001 = (energies_ncl[1] - energies_ncl[3])/N_FM E_110_001 = (energies_ncl[2] - energies_ncl[3])/N_FM msg = 'magnetocrystalline anisotropic energies are:' print(msg) writeLog(msg) msg = 'E[100]-E[001] = '+str(E_100_001*1e6)+' ueV/magnetic_atom' print(msg) writeLog(msg) msg = 'E[010]-E[001] = '+str(E_010_001*1e6)+' ueV/magnetic_atom' print(msg) writeLog(msg) msg = 'E[110]-E[001] = '+str(E_110_001*1e6)+' ueV/magnetic_atom' print(msg) writeLog(msg) if np.around(E_100_001,9) == np.around(E_010_001,9) and np.around(E_100_001,9) == np.around(E_110_001,9): TC_XY = ((final_list[1][1]-final_list[0][1])/(N*factor))*(0.89/(8*kB)) print(lcm_atoms) msg = 'this material seems to be an XY magnet, the calculated TC is '+str(TC_XY)+' k, not performing MC' print(msg) writeLog(msg) sys.exit() if not os.path.exists(root_path+'/input_MC'): msg = 'no input_MC file detected, writing this' print(msg) writeLog(msg) energies_xyz = [energies_ncl[0], energies_ncl[1], energies_ncl[3]] k_x = (energies_ncl[0]-np.max(energies_xyz))/(np.square(S_FM)*N_FM) k_y = (energies_ncl[1]-np.max(energies_xyz))/(np.square(S_FM)*N_FM) k_z = (energies_ncl[3]-np.max(energies_xyz))/(np.square(S_FM)*N_FM) T_MF = (S_FM*(S_FM+1)/(3*kB))*(J1*len(N1s)) + (S_FM*(S_FM+1)/(3*kB))*(J2*len(N2s)) + (S_FM*(S_FM+1)/(3*kB))*(J3*len(N3s)) + ( S_FM*(S_FM+1)/(3*kB))*(J4*len(N4s)) + (S_FM*(S_FM+1)/(3*kB))*(J5*len(N5s)) f = open('input_MC','w+') f.write('directory = MC_Heisenberg\n') f.write('repeat = 50 50 1\n') f.write('restart = 0\n') f.write('J1 (eV/link) = '+str(J1)+'\n') f.write('J2 (eV/link) = '+str(J2)+'\n') f.write('J3 (eV/link) = '+str(J3)+'\n') f.write('J4 (eV/link) = '+str(J4)+'\n') f.write('J5 (eV/link) = '+str(J5)+'\n') f.write('EMA = '+str(EMA)+'\n') f.write('k_x (eV/atom) = '+str(k_x)+'\n') f.write('k_y (eV/atom) = '+str(k_y)+'\n') f.write('k_z (eV/atom) = '+str(k_z)+'\n') f.write('T_start (K) = 1e-6\n') f.write('T_end (K) = '+str(T_MF)+'\n') f.write('div_T = 25\n') f.write('mu (mu_B/atom) = '+str(S_FM*2)+'\n') f.write('MCS = 100000\n') f.write('thresh = 10000\n') f.close() msg = 'successfully written input_MC, now will try to run Monte-Carlo based on this' print(msg) writeLog(msg) msg = 'if you want to run the MC with some other settings, make the neccesarry changes in input_MC and stop and re-run this script' print(msg) writeLog(msg) sleep(3) else: msg = 'existing input_MC detected, will try to run the MC based on this' print(msg) writeLog(msg) sleep(3) with open('input_MC') as f: for line in f: row = line.split() if 'directory' in line: path = root_path+'/'+row[-1] elif 'restart' in line: restart = int(row[-1]) elif 'repeat' in line: rep_z = int(row[-1]) rep_y = int(row[-2]) rep_x = int(row[-3]) elif 'J1' in line: J1 = np.double(row[-1]) elif 'J2' in line: J2 = np.double(row[-1]) elif 'J3' in line: J3 = np.double(row[-1]) elif 'J4' in line: J4 = np.double(row[-1]) elif 'J5' in line: J5 = np.double(row[-1]) elif 'k_x' in line: k_x = np.double(row[-1]) elif 'k_y' in line: k_y = np.double(row[-1]) elif 'k_z' in line: k_z = np.double(row[-1]) elif 'T_start' in line: Tstart = float(row[-1]) elif 'T_end' in line: Trange = float(row[-1]) elif 'div_T' in line: div_T = int(row[-1]) elif 'mu' in line: mu = float(row[-1]) elif 'MCS' in line: trange = int(row[-1]) elif 'thresh' in line: threshold = int(row[-1]) if os.path.exists(path): new_name = path+str(time()) os.rename(path,new_name) msg = 'found an old MC directory, renaming it to '+new_name print(msg) writeLog(msg) os.makedirs(path) repeat = [rep_x,rep_y,rep_z] S = mu/2 struc = Structure.from_file('2D.xsf') os.chdir(path) if restart==0: analyzer = CollinearMagneticStructureAnalyzer(struc,overwrite_magmom_mode='replace_all_if_undefined') struc_mag = analyzer.get_structure_with_only_magnetic_atoms() struc_mag.make_supercell(repeat) N = len(struc_mag) dr_max = ds[0] d_N1 = ds[1] d_N2 = ds[2] d_N3 = ds[3] d_N4 = ds[4] d_N5 = ds[5] all_coords = [0]*N for i in range(N): all_coords[i] = [struc_mag[i].x,struc_mag[i].y,struc_mag[i].z] all_coords = np.array(all_coords,dtype='float32') all_coords = all_coords.flatten() N1list = [[1,2]]*N N2list = [[1,2]]*N N3list = [[1,2]]*N N4list = [[1,2]]*N N5list = [[1,2]]*N if GPU_accel: nf = NfinderGPU msg = 'neighbor mapping will try to use GPU acceleration' print(msg) writeLog(msg) else: nf = Nfinder msg = 'neighbor mapping will be sequentially done in CPU, can be quite slow' print(msg) writeLog(msg) start_time_map = time() for i in range(N): N1list[i] = nf(struc_mag,i,d_N1,dr_max) if J2!=0: N2list[i] = nf(struc_mag,i,d_N2,dr_max) if J3!=0: N3list[i] = nf(struc_mag,i,d_N3,dr_max) if J4!=0: N4list[i] = nf(struc_mag,i,d_N4,dr_max) if J5!=0: N5list[i] = nf(struc_mag,i,d_N5,dr_max) print(str(i)+' / '+str(N-1)) if padding: msg = 'anticipating inhomogenous number of neighbors for some atoms, trying padding' print(msg) writeLog(msg) make_homogenous(N1list) make_homogenous(N2list) make_homogenous(N3list) make_homogenous(N4list) make_homogenous(N5list) end_time_map = time() time_map = np.around(end_time_map - start_time_map, 2) with open('N1list', 'wb') as f: dump(N1list, f) with open('N2list', 'wb') as f: dump(N2list, f) with open('N3list', 'wb') as f: dump(N3list, f) with open('N4list', 'wb') as f: dump(N4list, f) with open('N5list', 'wb') as f: dump(N5list, f) msg = 'neighbor mapping finished and dumped' print(msg) writeLog(msg) msg = 'the neighbor mapping process for a '+str(N)+' site lattice took '+str(time_map)+' s' print(msg) writeLog(msg) else: with open('N1list', 'rb') as f: N1list = load(f) with open('N2list', 'rb') as f: N2list = load(f) with open('N3list', 'rb') as f: N3list = load(f) with open('N4list', 'rb') as f: N4list = load(f) with open('N5list', 'rb') as f: N5list = load(f) N = len(N1list) print('neighbor mapping successfully read') N1list = np.array(N1list) N2list = np.array(N2list) N3list = np.array(N3list) N4list = np.array(N4list) N5list = np.array(N5list) temp = N1list.flatten() corrupt = np.count_nonzero(temp == -5) msg = 'the amount of site corruption in NNs is ' + str(corrupt) + ' / ' + str(len(temp)) + ', or ' + str(100.0*corrupt/len(temp)) + '%' print(msg) writeLog(msg) if J2!=0: temp = N2list.flatten() corrupt = np.count_nonzero(temp == -5) msg = 'the amount of site corruption in NNNs is ' + str(corrupt) + ' / ' + str(len(temp)) + ', or ' + str(100.0*corrupt/len(temp)) + '%' print(msg) writeLog(msg) if J3!=0: temp = N3list.flatten() corrupt = np.count_nonzero(temp == -5) msg = 'the amount of site corruption in NNNNs is ' + str(corrupt) + ' / ' + str(len(temp)) + ', or ' + str(100.0*corrupt/len(temp)) + '%' print(msg) writeLog(msg) if J4!=0: temp = N4list.flatten() corrupt = np.count_nonzero(temp == -5) msg = 'the amount of site corruption in NNNNNs is ' + str(corrupt) + ' / ' + str(len(temp)) + ', or ' + str(100.0*corrupt/len(temp)) + '%' print(msg) writeLog(msg) if J5!=0: temp = N5list.flatten() corrupt = np.count_nonzero(temp == -5) msg = 'the amount of site corruption in NNNNNNs is ' + str(corrupt) + ' / ' + str(len(temp)) + ', or ' + str(100.0*corrupt/len(temp)) + '%' print(msg) writeLog(msg) Ts = np.linspace(Tstart,Trange,div_T) Ms = [] Xs = [] start_time_mc = time() for T in Ts: spin_abs = S*np.ones(N) spin_x = np.zeros(N) spin_y = np.zeros(N) spin_z = S*np.ones(N) mags = np.zeros(trange-threshold) magsqs = np.zeros(trange-threshold) M,Msq=tFunc(spin_abs,spin_x,spin_y,spin_z,mags,magsqs,T,J2!=0,J3!=0,J4!=0,J5!=0) X = (Msq-np.square(M))/T Ms.append(M) Xs.append(X) print(str(T)+' '+str(M)+' '+str(X)) f = open(str(struc.formula).replace(' ','').replace('1','')+'_'+str(int(np.floor(Tstart)))+'-'+str(int(np.floor(Trange)))+'.dat','a+') f.write(str(T)+' '+str(M)+' '+str(X)+'\n') f.close() end_time_mc = time() time_mc = np.around(end_time_mc - start_time_mc, 2) msg = 'MC simulation have finished, analyse the output to determine the Curire temp.' print(msg) writeLog(msg) msg = 'the MC simulation took '+str(time_mc)+' s' print(msg) writeLog(msg) end_time_global = time() time_global = np.around(end_time_global - start_time_global, 2) msg = 'the whole end-to-end process took '+str(time_global)+' s' print(msg) writeLog(msg)
Python
2D
rpestourie/fdfd_local_field
src/simulation_unit_cell.jl
.jl
4,271
110
# TODO: modularize code, write simulate_unit_cell independent of the geometry, all geometry related code should be in geometry_code """ ```x, y, Ez, dpml, dsource, resolution = simulate_unit_cell(ps; refractive_indexes=zeros(3), frequency=1, interstice = 0.5, hole = 0.75, Ly = 17)``` This function simulates Helmholtz equation for a unit-cell consisting of holes in a substrate.""" function simulation_hole_layers_unit_cell(ps; refractive_indexes=zeros(3), frequency=1, interstice = 0.5, hole = 0.75, Ly = 17) #parameters for Maxwell solver # specific to the unit-cell Lx = 0.95 # Ly = 17 ω = 2pi*frequency dpml = 2 resolution = 40 dsource = 1 # specific to the unit-cell if refractive_indexes == zeros(3) refractive_indexes = Float64[1.0, 1.0, 1.45] end # ϵ_hole_layers is specific to the unit cell, all the rest is not A, nx, ny, x, y = Maxwell_2d(Lx, Ly, (x,y)-> ϵ_hole_layers(x, y, ps, refractive_indexes=refractive_indexes, interstice = interstice, hole = hole), ω, dpml, resolution) J = zeros(ComplexF64, (ny, nx)) J[end-(dpml + dsource) * resolution, :] = 1im * ω * ones(nx)*resolution Ez = reshape(A \ J[:], (ny, nx)) return x, y, Ez, dpml, dsource, resolution end """ ```function simulation_pillar_unit_cell(p; kwargs=Dict())``` This function simulated Helmholtz equation for a unit-cell consisting of a pillar on top of a substrate. @ parameters: - p: width of the pillar in adimensional units - kwargs: dictionary with parameters of the unit cell we would want to change @return (everything needed to compute the local field) - x, y: coordinates - Ez: values of the electric field out of plane - dpml - dsource is the distance of the source from the pml (in adimensional unit), also use for the distance between the monitor and the pml - resolution """ function simulation_pillar_unit_cell(p; kwargs=Dict()) @assert length(p) == 1 # parameter unit cell height= get(kwargs, "height", 600/443) #parameters for Maxwell solver refractive_indexes= get(kwargs, "refractive_indexes", [1., 2., 1.4]) frequency= get(kwargs, "frequency", 1.) Ly = get(kwargs, "Ly", 7) resolution = get(kwargs, "resolution", 40) dpml = get(kwargs, "dpml", 2) Lx = get(kwargs, "Lx", 1) dsource = get(kwargs, "dsource", 1) ω = 2pi*frequency # specific to the unit-cell A, nx, ny, x, y = Maxwell_2d(Lx, Ly, (x,y)-> ϵ_pillar_function(x, y, [p, height], refractive_indexes=refractive_indexes), ω, dpml, resolution) J = zeros(ComplexF64, (ny, nx)) J[end-(dpml + dsource) * resolution, :] = 1im * ω * ones(nx)*resolution Ez = reshape(A \ J[:], (ny, nx)) return x, y, Ez, dpml, dsource, resolution end """ ```x, y, zeroth_order = get_local_field(ps; refractive_indexes=zeros(3), frequency=1, interstice = 0.5, hole = 0.75, Ly = 17)``` this function simulates the structure with holes using Maxwell's equations, and computes the zeroth-order Fourier coefficient. input arguments: - ps: array with parametrization of the unit-cell corresponding the widths of holes in the substrate - refractive_indexes: optional argument with refractive indexes of background, hole and substrate. For reference simulation: set refractive indexes to ones(3)*eps_substrate - frequency (optional): frequency of simulation - interstice (optional): number of wavelength between holes - hole (optional): number of wavelength inside holes - Ly (optional): length of the computational domain returns: - return the complex zeroth order Fourier coefficient of the transmitted field """ function get_local_field(ps; simulate_unit_cell=simulation_hole_layers_unit_cell, refractive_indexes=zeros(3), frequency=1, interstice = 0.5, hole = 0.75, Ly = 17) x, y, Ez, dpml, dsource, resolution = simulate_unit_cell(ps, refractive_indexes=refractive_indexes, frequency=frequency, interstice = interstice, hole = hole, Ly = Ly) return x, y, mean(Ez[(dpml + dsource) * resolution, :]) end function get_local_field(ps, simulate_unit_cell; kwargs=Dict()) x, y, Ez, dpml, dsource, resolution = simulate_unit_cell(ps, kwargs=kwargs) return x, y, mean(Ez[(dpml + dsource) * resolution, :]) end
Julia
2D
rpestourie/fdfd_local_field
src/misc.jl
.jl
1,929
50
function rescaling(x, lb, ub) @assert all([x <= 1, x >= 0]) lb + x * (ub-lb) end function create_parameters(number_holes, number_training_data, δ, lb, ub) ps_vec = zeros(number_holes) for it_training = 1:number_training_data cur_ps = rescaling.(rand(number_holes), lb, ub) ps_vec = vcat(ps_vec, cur_ps) for it_hole = 1:number_holes perturb = zeros(number_holes) perturb[it_hole]+= δ ps_vec = vcat(ps_vec, cur_ps .+ perturb) end end return transpose(reshape(ps_vec[number_holes+1:end], (number_holes, number_training_data*6))) end function finite_difference_deriv(curv, vperturb, δ) return (vperturb-curv)/δ end function get_val_gradient_from_data(data_array; δ=1e-2) n_variable = size(data_array)[2]-2 n_data_val_gradient = size(data_array)[1]÷(1+n_variable) real_val_gradient = zeros(Float64, (n_data_val_gradient, 2*n_variable+1)) imag_val_gradient = zeros(Float64, (n_data_val_gradient, 2*n_variable+1)) for i=1:n_data_val_gradient real_vector_data = data_array[(i-1)*(1+n_variable)+1, 1:n_variable] #push the data input real_curval = data_array[(i-1)*(1+n_variable)+1, n_variable+1] push!(real_vector_data, real_curval) imag_vector_data = data_array[(i-1)*(1+n_variable)+1, 1:n_variable] imag_curval = data_array[(i-1)*(1+n_variable)+1, n_variable+2] push!(imag_vector_data, imag_curval) for j=1:n_variable push!(real_vector_data, finite_difference_deriv(real_curval, data_array[(i-1)*(1+n_variable)+1+j, n_variable+1], δ)) push!(imag_vector_data, finite_difference_deriv(imag_curval, data_array[(i-1)*(1+n_variable)+1+j, n_variable+2], δ)) end real_val_gradient[i,:] = real_vector_data imag_val_gradient[i,:] = imag_vector_data end return real_val_gradient, imag_val_gradient end
Julia
2D
rpestourie/fdfd_local_field
src/fdfd_maxwell_solver.jl
.jl
2,488
81
""" ```Maxwell_2d(Lx, Ly, ϵ, ω, dpml, resolution; Rpml=1e-20, show_geometry=false)``` This function compute the finite difference operator for a Maxwell Finite Difference Frequency Domain solver. The solver is in 2D and handles the out-of-plane polarization. Boundary conditions: - periodic in x - pml at y = Ly and y = 0 Argument: - Lx : length in x direction - Ly : length in y direction - ϵ : function that gives the permittivity in function of a spatial point - ω : frequency - dpml : width of the pml - resolution : number of points per unit - Rpml : pml extinction parameter - show_geometry : flag to plot the geometry Returns: - the operator matrix - the number of points in x direction - the number of points in y direction - x grid points - y grid points """ function Maxwell_2d(Lx, Ly, ϵ, ω, dpml, resolution; Rpml=1e-20) nx = round(Int, Lx * resolution) #nb point in x ny = round(Int, (Ly + 2*dpml) * resolution) #nb points in y npml = round(Int, dpml*resolution) δ = 1/resolution # coordinates centered in (0,0) x = (1:nx) * δ y = (1-npml:ny-npml) * δ #define the laplacian operator in x direction o = ones(nx)/δ # D = spdiagm((-o,o), (-1,0), nx+1, nx) # v0.6 Imat, J, V = SparseArrays.spdiagm_internal(-1 => -o, 0 => o); D = sparse(Imat, J, V, nx+1, nx) ∇2x = transpose(D) * D #periodic boundary condition in x direction ∇2x[end,1]-=1/δ^2 ∇2x[1,end]-=1/δ^2 #define the laplacian operator in y direction o = ones(ny) / δ σ0 = -log(Rpml) / (4dpml^3/3) y′=((-npml:ny-npml) .+ 0.5) * δ σ = Float64[ξ>Ly ? σ0 * (ξ-Ly)^2 : ξ<0 ? σ0 * ξ^2 : 0.0 for ξ in y] Σ = spdiagm(0 => 1.0 ./(1 .+ (im/ω)*σ)) σ′ = Float64[ξ>Ly ? σ0 * (ξ-Ly)^2 : ξ<0 ? σ0 * ξ^2 : 0.0 for ξ in y′] Σ′ = spdiagm(0 => 1.0 ./(1 .+ (im/ω)*σ′)) # D = spdiagm((-o, o), (-1, 0), ny+1, ny) # v0.6 Imat, J, V = SparseArrays.spdiagm_internal(-1 => -o, 0 => o); D = sparse(Imat, J, V, ny+1, ny) ∇2y = Σ * transpose(D) * Σ′ * D #get 2d laplacian using kronecker product Ix = sparse(1.0I, nx, nx) # Ix = speye(nx) # v0.6 Iy = sparse(1.0I, ny, ny) # Iy = speye(ny) # v0.6 ∇2d = (kron(Ix, ∇2y) + kron(∇2x, Iy)) # geometry = ComplexF64[ϵ(ξ, ζ) for ζ in y, ξ in x] geometry = ϵ(x, y) return (∇2d - spdiagm(0 => reshape(ω^2 * geometry, length(x)*length(y))), nx, ny, x, y) end
Julia
2D
rpestourie/fdfd_local_field
src/fdfd_local_field.jl
.jl
201
9
using Statistics using LinearAlgebra using SparseArrays include("../src/fdfd_maxwell_solver.jl") include("../src/geometry_code.jl") include("../src/simulation_unit_cell.jl") include("../src/misc.jl")
Julia
2D
rpestourie/fdfd_local_field
src/geometry_code.jl
.jl
6,132
174
""" ```ϵ_hole_layers(x, y, ps, interstice = 0.5, hole = 0.75)``` return the permittivity of a unit-cell which consists of air holes in silica Arguments: - ps : widths of the air holes (need to be unit-less) - refractive_indexes : optional argument with refractive indexes of background, hole and substrate. For reference simulation: set refractive indexes to ones(3)*eps_substrate - interstice (optional): number of length_scale in between holes acounting for the permittivity of the substrate - hole (optional): number of length_scale in the holes Returns: - geometry : a complex array with the epsilon data of the unit-cell """ function ϵ_hole_layers(x, y, ps; refractive_indexes=zeros(3), interstice = 0.5, hole = 0.75) @assert length(x)> 2 # makes sure δ is defined nx, ny = length(x), length(y) δ = x[2] - x[1] Ly_pml = y[end] - y[1] + δ Lx = x[end] - x[1] + δ @assert all(ps .> δ) # makes sure pixel-averaging handles all cases @assert all(ps .<= Lx) # makes sure that the holes are not bigger than the period # material properties of the unit-cell if refractive_indexes == zeros(3) refractive_index_background = 1.0 refractive_index_hole = 1.0 refractive_index_substrate = 1.45 else refractive_index_background, refractive_index_hole, refractive_index_substrate = refractive_indexes end eps_background, eps_hole, eps_substrate = refractive_index_background^2, refractive_index_hole^2, refractive_index_substrate^2 geometry = ones(ComplexF64, ny, nx) * eps_background index_top_substrate = floor(Int64, Ly_pml * 0.35/ δ) # 80% of the domain is substrate # substrate geometry[index_top_substrate:end, :] .= eps_substrate # handles case for sub-pixel averaging if x[nx÷2] == 0 w_offset=1/2 else w_offset=0 end # holes number_holes = length(ps) n_inter_hole = floor(Int64, interstice / refractive_index_substrate / δ) n_hole_height = floor(Int64, hole / δ) @assert index_top_substrate + number_holes * (n_inter_hole + n_hole_height) < ny # makes sure that Ly is big enough for the holes (possibly in PML) for it_holes = 1:number_holes half_width = ps[it_holes]/2δ - w_offset n_half_width = floor(Int64, half_width) weight_eps_hole = half_width - n_half_width # inside holes n_start = floor(Int64, (nx - 2*n_half_width)/2 - w_offset) geometry[index_top_substrate + it_holes * n_inter_hole + (it_holes-1) * n_hole_height + 1:index_top_substrate + it_holes * (n_inter_hole + n_hole_height), n_start + 1: n_start + floor(Int64, 2*(n_half_width + w_offset)) + 1] .= eps_hole # pixel averaging # left geometry[index_top_substrate + it_holes * n_inter_hole + (it_holes-1) * n_hole_height + 1:index_top_substrate + it_holes * (n_inter_hole + n_hole_height), n_start] .= weight_eps_hole * eps_hole + (1 - weight_eps_hole) * eps_substrate # right geometry[index_top_substrate + it_holes * n_inter_hole + (it_holes-1) * n_hole_height + 1:index_top_substrate + it_holes * (n_inter_hole + n_hole_height), end-n_start+1] .= weight_eps_hole * eps_hole + (1 - weight_eps_hole) * eps_substrate end return geometry end """ ```function ϵ_pillar_function(x, y, ps; refractive_indexes=zeros(3))``` return the permittivity of a unit-cell which consists of a pillar on top of a substrate Arguments: - ps : width and height of the pillar (adimensional) - refractive_indexes : optional argument with refractive indexes of background, hole and substrate. For reference simulation: set refractive indexes to ones(3)*eps_substrate Returns: - geometry : a complex array with the epsilon data of the unit-cell """ function ϵ_pillar_function(x, y, ps; refractive_indexes=zeros(3)) @assert length(x)> 2 # makes sure δ is defined @assert length(ps) == 2 width = ps[1] height = ps[2] nx, ny = length(x), length(y) δ = x[2] - x[1] Ly_pml = y[end] - y[1] + δ Lx = x[end] - x[1] + δ p = width/Lx @assert all(ps .> δ) # makes sure pixel-averaging handles all cases @assert (width <= Lx) # makes sure that the pillar width are not bigger than the period @assert (height <= Ly_pml) # makes sure that height fits in domain (TODO: need to be careful of PML) # material properties of the unit-cell if refractive_indexes == zeros(3) refractive_index_background = 1.0 refractive_index_pillar = 2.0 refractive_index_substrate = 1.4 else refractive_index_background, refractive_index_pillar, refractive_index_substrate = refractive_indexes end eps_background, eps_pillar, eps_substrate = refractive_index_background^2, refractive_index_pillar^2, refractive_index_substrate^2 geometry = ones(ComplexF64, ny, nx) * eps_background index_top_substrate = floor(Int64, Ly_pml * 0.5/ δ) # 80% of the domain is substrate # substrate geometry[index_top_substrate:end, :] .= eps_substrate # handles case for sub-pixel averaging if x[nx÷2] == 0 w_offset=1/2 else w_offset=0 end # pillar n_pillar_height = floor(Int64, height / δ) half_width = p/2δ - w_offset n_half_width = floor(Int64, half_width) weight_eps_hole = half_width - n_half_width # inside pillar n_start = floor(Int64, (nx - 2*n_half_width)/2 - w_offset) geometry[index_top_substrate-n_pillar_height:index_top_substrate-1, n_start + 1: n_start + floor(Int64, 2*(n_half_width + w_offset)) + 1] .= eps_pillar # pixel averaging # left geometry[index_top_substrate-n_pillar_height:index_top_substrate-1, n_start] .= weight_eps_hole * eps_pillar + (1 - weight_eps_hole) * eps_background # right geometry[index_top_substrate-n_pillar_height:index_top_substrate-1, end-n_start+1] .= weight_eps_hole * eps_pillar + (1 - weight_eps_hole) * eps_background return geometry end
Julia
2D
rpestourie/fdfd_local_field
test/test_get_gradient.jl
.jl
1,023
20
data_array = [0.405503 0.707334 0.211959 0.2539 0.479168 0.161135 -0.0888367 0.415503 0.707334 0.211959 0.2539 0.479168 0.153381 -0.102154 0.405503 0.717334 0.211959 0.2539 0.479168 0.14144 -0.106894 0.405503 0.707334 0.221959 0.2539 0.479168 0.150405 -0.0882103 0.405503 0.707334 0.211959 0.2639 0.479168 0.135277 -0.108619 0.405503 0.707334 0.211959 0.2539 0.489168 0.162739 -0.0966177] real_gd, imag_gd = get_val_gradient_from_data(data_array, δ=1e-2) @test imag_gd[1, end-4] == (data_array[2, end] - data_array[1, end])*1e2 @test imag_gd[1, end-2] == (data_array[4, end] - data_array[1, end])*1e2 @test imag_gd[1, end] == (data_array[6, end] - data_array[1, end])*1e2 @test real_gd[1, end-4] == (data_array[2, end-1] - data_array[1, end-1])*1e2 # note that this value is wrong because the data was simulated with a δ=1e-2 pertubation real_gd, imag_gd = get_val_gradient_from_data(data_array, δ=1e-4) @test imag_gd[1, end-4] ≈ (data_array[2, end] - data_array[1, end])*1e4
Julia
2D
rpestourie/fdfd_local_field
test/test_fdfd_maxwell_solver.jl
.jl
577
11
# checks Fresnel's transmission coefficient numerically n_incident = 1. n_transmit = 1.45 ref_local_field = get_local_field(Float64[0.5], refractive_indexes=ones(3) * n_incident)[3] interface_localfield = get_local_field(Float64[0.5], refractive_indexes=Float64[n_transmit, n_incident, n_incident])[3] transmission_numerical = abs2(interface_localfield)/abs2(ref_local_field)/n_incident*n_transmit transmission_fresnel = ((2n_incident)/(n_incident+n_transmit))^2/n_incident*n_transmit @test abs(transmission_fresnel - transmission_numerical)/abs(transmission_fresnel)< 1e-2
Julia
2D
rpestourie/fdfd_local_field
test/runtests.jl
.jl
409
10
using Test include("../src/fdfd_local_field.jl") @testset "fdfd_local_field" begin @testset "maxwell_fdfd_solver" begin include("test_fdfd_maxwell_solver.jl") end @testset "pixel averaging" begin include("test_geometry_code.jl") end @testset "finite difference gradient" begin include("test_get_gradient.jl") end @testset "parallelization" begin include("test_parallel_vs_serial.jl") end end
Julia
2D
rpestourie/fdfd_local_field
test/test_parallel_vs_serial.jl
.jl
970
27
number_training_data = 6 r = rand() δ = 1e-2 number_holes = 5 lb = 0.1 # for a blue wavelength this corresponds to a minimum feature of ≈40 nm ub = 0.95 - lb # this assumes a period of 0.95 λ # create parameters ps_array = create_parameters(number_holes, number_training_data, δ, lb, ub) # reference simulation refractive_indexes_ref = ones(3) * 1.45 ref_local_field = get_local_field(Float64[0.5], refractive_indexes=refractive_indexes_ref)[3] # simulations refractive_indexes_sim = Float64[1.0, 1.0, 1.45] sim_local_fields = ComplexF32[get_local_field(ps_array[it_ps, :], refractive_indexes=refractive_indexes_sim)[3] for it_ps=1:size(ps_array)[1]] sim_local_fields2 = zeros(ComplexF64, (number_holes+1)*number_training_data) @inbounds Base.Threads.@threads for it_ps=1:size(ps_array)[1] sim_local_fields2[it_ps] = get_local_field(ps_array[it_ps, :], refractive_indexes=refractive_indexes_sim)[3] end @test all(sim_local_fields .≈ sim_local_fields2)
Julia
2D
rpestourie/fdfd_local_field
test/test_geometry_code.jl
.jl
490
15
ps = [37/40] refractive_indexes = Float64[1., 1., sqrt(3)] x, y, Ez, dpml, dsource, resolution = simulation_hole_layers_unit_cell(ps, refractive_indexes=refractive_indexes) δ = x[2] - x[1] Ly_pml = y[end] - y[1] + δ # test that the unit-cell is correct @test resolution == 40 @test Ly_pml == 16 # test that some permittivity are 2 which is the desired pixel-averaging geometry = real.(ϵ_hole_layers(x, y, ps, refractive_indexes=refractive_indexes)) @test any(isapprox.(geometry, 2.))
Julia
2D
rpestourie/fdfd_local_field
examples/example_Ez_geometry_plots.jl
.jl
805
25
include("../src/fdfd_local_field.jl") using Plots pyplot() ### reference simulation refractive_indexes = ones(3) * 1.45 ps = [0.75, 0.5, 0.66, 0.33, 0.66, 0.5, 0.75, 0.5, 0.66, 0.33] x, y, Ez, dpml, dsource, resolution = simulation_hole_layers_unit_cell(ps, refractive_indexes=refractive_indexes, frequency=0.25) heatmap(real.(ϵ_hole_layers(x, y, ps, refractive_indexes=refractive_indexes))) gui() heatmap(real.(Ez), c=ColorGradient([:red, :white, :blue])) gui() ### structure simulation refractive_indexes = Float64[1.0, 1.0, 1.45] x, y, Ez, dpml, dsource, resolution = simulation_hole_layers_unit_cell(ps, refractive_indexes=refractive_indexes) heatmap(real.(ϵ_hole_layers(x, y, ps, refractive_indexes=refractive_indexes))) gui() heatmap(real.(Ez), c=ColorGradient([:red, :white, :blue])) gui()
Julia
2D
rpestourie/fdfd_local_field
examples/serial_simulations.jl
.jl
1,219
35
include("../src/fdfd_local_field.jl") cd("examples/") using BenchmarkTools using DelimitedFiles println("$(Threads.nthreads()) threads!") number_training_data = 150000 # ideally 1e5 here r = rand() δ = 1e-2 fname = "test_data_$(round(Int, r*1e5))_$δ.csv" number_holes = 10 lb = 0.1 # for a blue wavelength this corresponds to a minimum feature of ≈40 nm ub = 0.95 - lb # this assumes a period of 0.95 λ # create parameters ps_array = create_parameters(number_holes, number_training_data, δ, lb, ub) # reference simulation refractive_indexes_ref = ones(3) * 1.45 ref_local_field = get_local_field(Float64[0.5], refractive_indexes=refractive_indexes_ref)[3] # simulations refractive_indexes_sim = Float64[1.0, 1.0, 1.45] # embarassingly parallel loop (used to be serial hence the name of the file) sim_local_fields = zeros(ComplexF64, (number_holes+1)*number_training_data) @time @inbounds Base.Threads.@threads for it_ps=1:size(ps_array)[1] sim_local_fields[it_ps] = get_local_field(ps_array[it_ps, :], refractive_indexes=refractive_indexes_sim)[3] end local_field_data = sim_local_fields./ref_local_field writedlm("./data/"*fname,hcat(ps_array, real.(local_field_data), imag.(local_field_data)), ',')
Julia
2D
rpestourie/fdfd_local_field
examples/Simulate_RGB_data.ipynb
.ipynb
63,282
113
{ "cells": [ { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "┌ Warning: PyPlot is using tkagg backend, which is known to cause crashes on MacOS (#410); use the MPLBACKEND environment variable to request a different backend.\n", "└ @ PyPlot /Users/raphaelpestourie/.julia/packages/PyPlot/XHEG0/src/init.jl:192\n" ] }, { "data": { "text/plain": [ "get_val_gradient_from_data (generic function with 1 method)" ] }, "execution_count": 1, "metadata": {}, "output_type": "execute_result" } ], "source": [ "using PyPlot\n", "include(\"../src/fdfd_local_field.jl\")" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [], "source": [ "wavelengths = Float64[488, 532, 658]\n", "lb = .15\n", "ub = .85\n", "\n", "### simulations\n", "# reference simulation\n", "refractive_indexes_ref = ones(3) * 1.45\n", "ref_local_field = Dict()\n", "for j in eachindex(wavelengths)\n", " ref_local_field[wavelengths[j]] = get_local_field(Float64[0.5], refractive_indexes=refractive_indexes_ref, frequency=wavelengths[j]/wavelengths[1])[3]\n", "end\n", "\n", "# simulations\n", "refractive_indexes_sim = Float64[1.0, 1.0, 1.45]\n", "ps = .15:0.01:0.85\n", "local_field_data = zeros(ComplexF64, (length(ps), length(wavelengths)))\n", "\n", "for (j, lambda) in enumerate(wavelengths)\n", " for (i, p) in enumerate(ps)\n", " freq = wavelengths[1]/wavelengths[j]\n", " local_field_data[i,j] = get_local_field([p], refractive_indexes=refractive_indexes_sim, frequency=freq)[3]/ref_local_field[wavelengths[j]]\n", " end\n", "end" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [ { "data": { "image/png": "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", "text/plain": [ "Figure(PyObject <Figure size 640x480 with 2 Axes>)" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "subplot(121)\n", "plot(angle.(local_field_data))\n", "subplot(122)\n", "plot(abs.(local_field_data));\n", "tight_layout()" ] }, { "cell_type": "code", "execution_count": 20, "metadata": {}, "outputs": [], "source": [ "# writedlm(\"RGB_uc_data.csv\", hcat(ps, real.(local_field_data), imag.(local_field_data)), ',')" ] } ], "metadata": { "kernelspec": { "display_name": "Julia 1.4.2", "language": "julia", "name": "julia-1.4" }, "language_info": { "file_extension": ".jl", "mimetype": "application/julia", "name": "julia", "version": "1.4.2" } }, "nbformat": 4, "nbformat_minor": 4 }
Unknown
2D
rpestourie/fdfd_local_field
examples/example_parameter_sweep_single_hole.jl
.jl
271
16
include("../src/fdfd_local_field.jl") using Plots pyplot() ps = transpose(0.1:0.01:0.9) E_local_field = [get_local_field(ps[:,i])[3] for i=1:length(ps)] Plots.plot(real.(E_local_field)) Plots.plot!(imag.(E_local_field)) gui() Plots.plot(angle.(E_local_field)) gui()
Julia
2D
rpestourie/fdfd_local_field
examples/get_val_gradients.jl
.jl
454
14
include("../src/fdfd_local_field.jl") using DelimitedFiles fname="examples/data/test_data__0.01.csv" data_array = readdlm(fname, ',') start_index = findnext('.', fname, 1) - 1 end_index = findnext('.', fname, start_index+2) - 1 δ = parse(Float64, fname[start_index:end_index]) real_gd, imag_gd = get_val_gradient_from_data(data_array, δ=δ) writedlm(fname[1:end-4]*"_real_gd.csv", real_gd, ',') writedlm(fname[1:end-4]*"_imag_gd.csv", imag_gd, ',')
Julia