    Next: Finding the root of Up: The Trust-Region subproblem Previous: must be positive definite.   Contents

Subsections

# Explanation of the Hard case.

Theorem 2 tells us that we should be looking for solutions to 4.9 and implicitly tells us what value of we need. Suppose that has an eigendecomposition: (4.8)

where is a diagonal matrix of eigenvalues and is an orthonormal matrix of associated eigenvectors. Then (4.9)

We deduce immediately from Theorem 2 that the value of we seek must satisfy (as only then is positive semidefinite) ( is the least eigenvalue of ). We can compute a solution for a given value of using: (4.10)

The solution we are looking for depends on the non-linear inequality . To say more we need to examine in detail. For convenience we define . We have that: (4.11)

where is , the component of .

## Convex example.

Suppose the problem is defined by: We plot the function in Figure 4.1. Note the pole of at the negatives of each eigenvalues of . In view of theorem 2, we are only interested in . If , the optimum lies inside the trust region boundary. Looking at the figure, we obtain , for . So, it means that if , we have an internal optimum which can be computed using 4.9. If , there is a unique value of (given in the figure and by (4.12)

which, used inside 4.9, give the optimal . ## Non-Convex example.

Suppose the problem is defined by: We plot the function in Figure 4.2. Recall that is defined as the least eigenvalue of . We are only interested in values of , that is . For value of , we have NOT positive definite. This is forbidden due to theorem 2. We can see that for any value of , there is a corresponding value of . Geometrically, H is indefinite, so the model function is unbounded from below. Thus the solution lies on the trust-region boundary. For a given , found using 4.14, we obtain the optimal using 4.9. ## The hard case.

Suppose the problem is defined by: We plot the function in Figure 4.3. Again, , is forbidden due to theorem 2. If, , there is no acceptable value of . This difficulty can only arise when is orthogonal to the space , of eigenvectors corresponding to the most negative eigenvalue of . When , then equation 4.9 has a limiting solution , where .  is positive semi-definite and singular and therefore 4.9 has several solutions. In particular, if is an eigenvector corresponding to , we have , and thus: (4.13)

holds for any value of the scalar . The value of can be chosen so that . There are two roots to this equation: and . We evaluate the model at these two points and choose as solution , the lowest one.    Next: Finding the root of Up: The Trust-Region subproblem Previous: must be positive definite.   Contents
Frank Vanden Berghen 2004-04-19