Documentation |
In the example Nonlinear Equations with Analytic Jacobian, the function nlsf1 computes the Jacobian J, a sparse matrix, along with the evaluation of F. What if the code to compute the Jacobian is not available? By default, if you do not indicate that the Jacobian can be computed in nlsf1 (by setting the Jacobian option in options to 'on'), fsolve, lsqnonlin, and lsqcurvefit instead uses finite differencing to approximate the Jacobian.
In order for this finite differencing to be as efficient as possible, you should supply the sparsity pattern of the Jacobian, by setting JacobPattern to a sparse matrix Jstr in options. That is, supply a sparse matrix Jstr whose nonzero entries correspond to nonzeros of the Jacobian for all x. Indeed, the nonzeros of Jstr can correspond to a superset of the nonzero locations of J; however, in general the computational cost of the sparse finite-difference procedure will increase with the number of nonzeros of Jstr.
Providing the sparsity pattern can drastically reduce the time needed to compute the finite differencing on large problems. If the sparsity pattern is not provided (and the Jacobian is not computed in the objective function either) then, in this problem with 1000 variables, the finite-differencing code attempts to compute all 1000-by-1000 entries in the Jacobian. But in this case there are only 2998 nonzeros, substantially less than the 1,000,000 possible nonzeros the finite-differencing code attempts to compute. In other words, this problem is solvable if you provide the sparsity pattern. If not, most computers run out of memory when the full dense finite-differencing is attempted. On most small problems, it is not essential to provide the sparsity structure.
Suppose the sparse matrix Jstr, computed previously, has been saved in file nlsdat1.mat. The following driver calls fsolve applied to nlsf1a, which is nlsf1 without the Jacobian. Sparse finite-differencing is used to estimate the sparse Jacobian matrix as needed.
function F = nlsf1a(x) % Evaluate the vector function n = length(x); F = zeros(n,1); i = 2:(n-1); F(i) = (3-2*x(i)).*x(i)-x(i-1)-2*x(i+1) + 1; F(n) = (3-2*x(n)).*x(n)-x(n-1) + 1; F(1) = (3-2*x(1)).*x(1)-2*x(2) + 1;
xstart = -ones(1000,1); fun = @nlsf1a; load nlsdat1 % Get Jstr options = optimoptions(@fsolve,'Display','iter','JacobPattern',Jstr,... 'Algorithm','trust-region-reflective','PrecondBandWidth',1); [x,fval,exitflag,output] = fsolve(fun,xstart,options);
In this case, the output displayed is
Norm of First-order Iteration Func-count f(x) step optimality 0 5 1011 19 1 10 16.0839 7.92496 1.92 2 15 0.0458179 1.3279 0.579 3 20 0.000101184 0.0631896 0.0203 4 25 3.16616e-07 0.00273698 0.00079 5 30 9.72483e-10 0.00018111 5.82e-05 Equation solved, inaccuracy possible. The vector of function values is near zero, as measured by the default value of the function tolerance. However, the last step was ineffective.
Alternatively, it is possible to choose a sparse direct linear solver (i.e., a sparse QR factorization) by indicating a "complete" preconditioner. For example, if you set PrecondBandWidth to Inf, then a sparse direct linear solver is used instead of a preconditioned conjugate gradient iteration:
xstart = -ones(1000,1); fun = @nlsf1a; load nlsdat1 % Get Jstr options = optimoptions(@fsolve,'Display','iter','JacobPattern',Jstr,... 'Algorithm','trust-region-reflective','PrecondBandWidth',inf); [x,fval,exitflag,output] = fsolve(fun,xstart,options);
and the resulting display is
Norm of First-order Iteration Func-count f(x) step optimality 0 5 1011 19 1 10 15.9018 7.92421 1.89 2 15 0.0128161 1.32542 0.0746 3 20 1.73502e-08 0.0397923 0.000196 4 25 1.10716e-18 4.55495e-05 2.74e-09 Equation solved. fsolve completed because the vector of function values is near zero as measured by the default value of the function tolerance, and the problem appears regular as measured by the gradient.
When the sparse direct solvers are used, the CG iteration is 0 for that (major) iteration, as shown in the output under CG-Iterations. Notice that the final optimality and f(x) value (which for fsolve, f(x), is the sum of the squares of the function values) are closer to zero than using the PCG method, which is often the case.