chjprod gives incorrect result when goth = .false.

[jfowkes]

When calling chjprod on ALLINITC with the default starting point x and Lagrange multipliers y:

double y0 = 3.5;
double p[] = {0.3, 0.3, 0.3, 0.3};
CUTEST_chjprod(&status, &CUTEst_nvar, &CUTEst_ncon, true, x, &y0, y, p, r);
CUTEST_cdh(&status, &CUTEst_nvar, &CUTEst_ncon, x+1.5, y-0.2, CUTEst_nvar, r);
CUTEST_chjprod(&status, &CUTEst_nvar, &CUTEst_ncon, true, NULL, &y0, NULL, p, r);
CUTEST_chjprod(&status, &CUTEst_nvar, &CUTEst_ncon, true, x+1.5, &y0, y-0.2, p, r);

gives

chjprod: [  4.2   2.1   8.4 -10.5]
chjprod: [ 94.314672   100.74225564 128.62426465  78.6433218 ]
chjprod: [330.40135202 352.89789475 450.18492627 275.25162631]

which is clearly incorrect.

[nimgould]

There must be something wrong with the setup, above, as if p is the zero vector, Hp = 0. This is what the fortran tests of the above say

[jfowkes]

Yes sorry p should be [0.3 0.3 0.3 0.3]

[nimgould]

The issue is that the call to cdh evaluates the Hessian of the Lagrangian, not the John function, and thus when calling chjprod again, we have the wrong Hessian. I will update the docs, and (perhaps) at some stage add a cdhj function