Note: Problem sets will be very similar to those from Spring, 2002.
If you care to work ahead, please see the course home
page for 2002
Problem Set # 1: Numerical integration --- potential energy terms
in atoms [HTMLPostscriptPDF ]
Solutions: [Postscript,
PDF].
Problem Set # 2: Solution of ODE's --- Poisson's equation in atoms and a simple Schrondinger solver [HTMLPostscriptPDF ]
Solutions: [Postscript,
PDF].
Problem Set # 3: Root finding -- Automated, high precision
Schrodinger solver for energies of atomic states [HTMLPostscriptPDF ]
Solutions: [Postscript,
PDF].
Problem Set # 4: Matching method; packaging -- Automated, high precision Schrodinger solver for wave functions of atoms; full set of routines needed for self-consistent atomic calculations [HTMLPostscriptPDF ]
Solutions: [Postscript,
PDF].
Problem Set # 5: Exchange and correlation energy functionals;
implementation of self-consistency;
density functional theory of atoms [HTMLPostscriptPDF ]
Solutions: [Postscript,
PDF].
Problem Set # 6: Optimization of atomic code, FLOP counting of inner
loop operations, cache optimization, ATLAS LAPACK, Octave -- [HTMLPostscriptPDF ]
Solutions: [Postscript,
PDF].
Problem Set # 7: Spectral solution of PDE's; expressive software:
Poisson's equation in a single line of octave code. [HTMLPostscriptPDF ]
Problem Set # 8 (Due April 29): Solution of Schrodinger's equation
in three dimensions using spectral methods and steepest descents
[HTMLPostscriptPDF ]
Problem Set # 9 (Due May 6): line minization, preconditioning,
conjugate gradientsl computation of the bond-length of the
hydrogen molecule, with your
own code built from scratch!!! [HTMLPostscriptPDF ]
Errata for Problem Set #9
The form for the
effective potential in Eq. (9) should be
Vsc=gbl_Vdual +cJdag(O(phi)) +cJdag(O(cJ(excVWN(n)))) +excpVWN(n).*cJdag(O(cJ(n)));
The final form for the total energy in Eq. (7) should be
-gbl_f*0.5*trace(Y'*L(Y)) +gbl_Vdual'*n +0.5*n'*cJdag(O(phi)) +cJ(excVWN(n))'*O(cJ(n));
The formula for the Polak-Ribiere form should be
beta=real(trace((g-og)'*K(g)))/real(trace((og-g)'*K(og)));
The formula for beta in the
Hestenes-Stiefel version of conjugate-gradients had a factor
of "K" in the denominator which should not be there. Please
re-down load the problem set (You may have to hit "reload" on your
browser in case the file is buffered.) to see the proper formula or
simply delete the "K" in the denominator only.