function swspace2d(method,scalem) % SWSPACE2D(method,scalem) % % FIGURE 3.1 of SIMONS & WANG, Int. J. Geomath. (2011) % % Spatial rendition of the radial part of the 2D-disk tapers scaled to % the unit interval. Checks that the finite-length effects are not so % grave by ensuring that the "small" Fourier components could be "zero" % after all. % % INPUT: % % method 'SE' by Slepian "extension" [default] % 'GL' by direct Gauss-Legendre integration % scalem 1 Scales solution for weightless orthogonality % 0 Scales solution for orthogonality with weight x [default] % % SEE ALSO: % % SWDISK, SWSPECTRAL2D, SDWSPACE, SDWSPECTRAL % % Last modified by fjsimons-at-alum.mit.edu, 09/20/2011 % The method of computation defval('method','SE'); % The method of scaling, if 0 don't do sqrt(x) scaling of Slepian defval('scalem',0); % Number of tapers per order NM=4; % Maximum order considered M=4; % Shannon number in question N=42; % The abscissas - as long as you like to display though see the % wavenumber-domain representation and SWSPECTRAL2D for the implications % of changing these values x=linspace(0,5,2^12); % This only if you do anything in the wavenumber domain [fax,selekt]=fftaxis1D(x,length(x),max(x)); kax=2*pi*fax; % Limiting wavenumber in question K=2*sqrt(N); % Bandlimiting array on the full set of wavenumbers selK=abs([kax ; -flipud(kax(2:end-1))])>K; % Start the computation and the figure clf [ah,ha,H]=krijetem(subnum(M+1,NM)); for m=0:M % The above are the unscaled functions "phi" or thus "g" [E1,V1{m+1},Nm,c,x,E2]=swdisk(m,N,NM,[],x,method,scalem); % This is the integral over the restricted interval for inds=1:length(V1{m+1}) werisit=x(:)<=1 & ~isnan(E1(:,inds)); difer(2*pi*trapeze(x(werisit),... x(werisit)'.*E1(werisit,inds).^2)-V1{m+1}(inds)) end E1m(m+1)=max(abs(E1(:))); if strcmp(method,'GL') && scalem==1 % Calculate the truly bandlimited spectrum FE1=fft(E1); FE1(selK,:)=0; % And calculate the truly bandlimited eigenfunctions E1F=ifft(FE1); % If the input was Hermitian you're all set difer(imag(E1F)) % Turns out in this case you almost get the input back - but only % when the normalization is unweighted, if not you'd probably have to % put the sqrt(x) in there with the ifft somehow. end % Plot the space-domain solution for ondex=1:NM axes(ah(m*NM+ondex)) pm(m+1,ondex)=plot(x,E1(:,ondex),'-','Color',grey(6)); hold on pg(m+1,ondex)=plot(x(x<=1),E1(x<=1,ondex),'-','Color','k'); if strcmp(method,'GL') && scalem==1 % Plot the ifft with the small components removed pgg(m+1,ondex)=plot(x,E1F(:,ondex),'-','Color','r'); end drawnow end end % Cosmetics set(ah,'xlim',[0 max(x)],'xtick',[0 1 max(x)],'xgrid','on','ygrid','on',... 'ylim',[-1.2 1.2]*max(E1m)) nolabels(ah(1:end-NM),1) nolabels(ha(M+2:end),2) longticks(ah,1/2) % Eigenvalue labels nf=9; lox={'ll','ur','lr','ul',... 'll','ur','lr','ul',... 'll','ur','lr','ul',... 'll','ur','lr','ul',... 'll','ur','lr','ul'}; for m=0:M for ondex=1:NM axes(ah(m*NM+ondex)) t{m+1,ondex}=sprintf('%s = %9.6f','\lambda',V1{m+1}(ondex)); if V1{m+1}(ondex)<0.975 lox='lr'; else lox='ur'; end [bh(m+1,ondex),th(m+1,ondex)]=... boxtex(lox,ah(m*NM+ondex),t{m+1,ondex},nf,[],0.75); end end set(th,'FontS',nf-1) for index=NM*M+1:NM*(M+1) axes(ah(index)) xl(index)=xlabel(sprintf('radius %s=r/R','\xi')); end for index=1:M+1 axes(ha(index)) yl(index)=ylabel(sprintf('m = %i',index-1)); end serre(H,1/3,'across') % So here this goes wrong in the second version for index=1:NM axes(ah(index)) tlb(index)=title(sprintf('%s = %i','\alpha',index)); end set([pm(:) ; pg(:)],'LineW',1) set(ah,'Fonts',nf) fig2print(gcf,'portrait') % set(gcf,'Color','w','Inv','off') shrink(ah,1,1/1.1) if strcmp(method,'GL') && scalem==1 set([pgg(:)],'LineW',0.5) q=input('Delete red lines? [y] ','s'); defval('q','y') if strcmp(q,'y') delete(pgg) end end figdisp([],[],[],0)