topic Using det in nlmixed in Statistical Procedures

Using det in nlmixed

user_123221 — Fri, 06 Oct 2017 07:58:41 GMT

Hi,

there is a procedure to calculate the determinant and I want to use it within the nlmixed procedure. To give an example I modified an nlmixed example from the doc. This doesn't make much sense but illustrates my point. The following works:

proc nlmixed data=theoph;
   parms beta1=-3.22 beta2=0.47 beta3=-2.45
         s2b1 =0.03  cb12 =0    s2b2 =0.4 s2=0.5;
   cl   = exp(beta1 + b1);
   ka   = exp(beta2 + b2);
   ke   = exp(beta3);
   pred = cl*beta1 - ka*ke;
   model conc ~ normal(pred,s2);
   random b1 b2 ~ normal([0,0],[s2b1,cb12,s2b2]) subject=subject;
run;

But when I substitute "cl*beta1 - ka*ke" by a seemingly equivalent call to det I get an error.

proc nlmixed data=theoph;
   parms beta1=-3.22 beta2=0.47 beta3=-2.45
         s2b1 =0.03  cb12 =0    s2b2 =0.4 s2=0.5;
   cl   = exp(beta1 + b1);
   ka   = exp(beta2 + b2);
   ke   = exp(beta3);
   array A [2,2];
   A[1,1] = cl;
   A[1,2] = ka;
   A[2,1] = ke;
   A[2,2] = beta1;
   call det(A, pred);
   model conc ~ normal(pred,s2);
   random b1 b2 ~ normal([0,0],[s2b1,cb12,s2b2]) subject=subject;
run;

The error is:

NOTE: Array argument A is ignored for the derivative of the 'DET' function.
ERROR: Random effect b1 not found in the model.

Is there a way to overcome this? Of course in my real code I want to calculate a larger determinant and also use other matrix operations.

Thanks

Re: Using det in nlmixed

Rick_SAS — Fri, 06 Oct 2017 14:09:41 GMT

I've not used the matrix capabilities of PROC NLMIXED, but I think the problem is that when you assign the elements of A by using A[1,1]=c1, etc. The program needs to know how to take the derivative of DET(A), and to do that it must know the symbols for each element of A. Try the following syntax, which associates the elements of matrix A to symbols in row-major order:

array A [2,2] cl ka ke beta1;

Re: Using det in nlmixed

user_123221 — Mon, 09 Oct 2017 06:56:51 GMT

When I specify the array in the way you suggested the "NOTE" message disappears but I still get

ERROR: Random effect b1 not found in the model.