# HP Forums

Full Version: LDLt decomposition?
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
hi all,
in HP Prime we have for a matrix LU decomposition (it give P, L, U so PA=LU) and Cholesky decomposition (for that A=LLt, Lt is TRAN(L))...
I can't find a factorization LDLt where D is a diagonal of pivot.
Is there a way to get it? a program or something?

Thank you!

Salvo
I wonder still for this question and another one, connected (in my case):
With pivot(Mat, n,m) we get pivot for n,m: what's a simple way to get every pivots (Gauss' reduction); I forgot it thanks!
Salvo

Code:
``` EXPORT gaussJordan(m) // Gauss-Jordan elimination and pivots // Salvo Micciché 2015 BEGIN local temp, temp2, gj, r, c, j, piv; r:=rowDim(m); c:=colDim(m); temp:=MAKEMAT(0,r,c); gj:=MAKEMAT(0,r,c); piv:=MAKELIST(0,X,1,r); gj(1):= m(1); piv(1):=gj(1,1);     temp:= pivot(m,1,1); FOR j FROM 2 TO r DO     temp:=delrows(temp,1);     temp:=delcols(temp,1);     temp:=temp/piv(j-1);     temp:= pivot(temp,1,1);     piv(j):=temp(1,1);     gj(j):= temp(1); END; // for RETURN {gj, piv}; END;```

but I get wrong results (see image): I should get 0 below pivots at left, not ...at right; I'm trying SWAPROW() but it doesn't run...
Any help?
Is diag() useful for you?

Code:
``` EXPORT gjp() BEGIN    M5:=[[2,1,1,5],[-8,-2,-12,0],[1,2,0,0]];  // Your matrix   M6:=diag(M5);  //  Diagonal Matrix   return M6;   END;```
(06-02-2015 11:26 AM)DrD Wrote: [ -> ]Is diag() useful for you?

Code:
``` EXPORT gjp() BEGIN    M5:=[[2,1,1,5],[-8,-2,-12,0],[1,2,0,0]];  // Your matrix   M6:=diag(M5);  //  Diagonal Matrix   return M6;   END;```

hi Dale,
diag() is also ok to get pivots, but after to have "pivot-ized" the original matrix: [[2,1,1,5],[4,-6,0,-2],[-2,7,2,9]]

My goal is to return a transformed matrix and a list with pivots (and also diag() could be ok).

EDIT:
Something like that, Dale:
Code:
``` EXPORT gaussJordan(m) // Gauss-Jordan elimination and pivots // Salvo Micciché 2015 BEGIN local temp, k, gj, r, c, j, piv; r:=rowDim(m); c:=colDim(m); temp:=MAKEMAT(0,r,c); gj:=MAKEMAT(0,r,c); piv:=MAKELIST(0,X,1,r); temp:= m; FOR j FROM 1 TO r DO       temp:=pivot(temp,1,1);       // gj(j):= temp(1);     FOR k FROM 1 TO colDim(temp) DO         gj(j, c-k+1):= temp(1, colDim(temp)-k+1);     END; // inner for       piv(j):=temp(1,1);     IF (j<r) THEN     temp:= delrows(temp,1);     temp:= delcols(temp,1);     END;       temp:= temp/piv(j);   END; // for piv:= list2mat(piv, r); RETURN {gj, piv}; END;```

this runs, but with "singular" matrices it gives still division by 0 and in some cases fails (we need also swapping of some rows...), however now we have a base to calculate LDLt decomposition also...
SWAPROW runs well in command line but not in a program? or am I missing something?

This code doesn't swap row 2 with 3...

Code:
```EXPORT gaussJordan(m) // Gauss-Jordan elimination and pivots // Salvo Micciché 2015 BEGIN local k, gj, r, c, j, piv; r:=rowDim(m); c:=colDim(m); gj:=MAKEMAT(0,r,c); piv:=MAKELIST(0,X,1,r); M1:= m; FOR j FROM 1 TO r DO       M1:=pivot(M1,1,1);       piv(j):=M1(1,1);     IF (piv(j)==0) THEN  PRINT("M1 first " + M1);     M1:=m; PRINT("M1 then " + M1);     SWAPROW(M1, j, (j+1));  PRINT("M1 and after " + M1);     CONTINUE;      END; // if     FOR k FROM 1 TO colDim(M1) DO         gj(j, c-k+1):= M1(1, colDim(M1)-k+1);     END; // inner for     IF (j<r) THEN     M1:= delrows(M1,1);     M1:= delcols(M1,1);     END;     IF piv(j)==0 THEN RETURN "Pivot is 0, division by 0"; END;       M1:= M1/piv(j);   END; // for piv:= list2mat(piv, r); RETURN {gj, piv}; END;```

See the prints in Terminal (debug, hi) with this matrix
[[1,2,3],[2,4,3],[3,2,1]]
Normally it would give division by zero, so we must swap the row with its next...
The result should be [[1,2,3],[0,-4,-10],[0,0,-3]] with pivots d={1,-4,-3}

I'm using an "old C style" tip, and it works:
Code:
```  // M1:= SWAPROW(M1, numpivot, (numpivot+1));  temp:= M1(numpivot); M1(numpivot):=M1(numpivot+1); M1(numpivot+1):=temp;```
I tried both SWAPROW() and SWAPCOL() on a 3x3 matrix and neither worked from the command line or HOME program, for me.

-Dale-
(06-02-2015 10:05 PM)DrD Wrote: [ -> ]I tried both SWAPROW() and SWAPCOL() on a 3x3 matrix and neither worked from the command line or HOME program, for me.

-Dale-

Please, could I suspect a bug? Strange, however...

Salvo
ALL similar matrix commands have the same requirement: ADDROW(), DELROW(), SWAPROW(), etc., all have the same CAS requirement in order to work on the HC and virtual calc. They seem to work ok on the Android app. I sent a bug report in, but, maybe this is intended behavior? If the command names were in lower case, it would serve as a hint that they need CAS handling.

-Dale-
(06-03-2015 11:59 AM)DrD Wrote: [ -> ]ALL similar matrix commands have the same requirement: ADDROW(), DELROW(), SWAPROW(), etc., all have the same CAS requirement in order to work on the HC and virtual calc. They seem to work ok on the Android app. I sent a bug report in, but, maybe this is intended behavior? If the command names were in lower case, it would serve as a hint that they need CAS handling.

-Dale-

I agree.
Here I've issues with them in CAS, however, so I think maybe there could be some bugs with the last FW upgrade...

Salvo

EDIT:
here SWAPROW([[1,2,3],[0,0,-3],[0,-4,-10]],2,3) doesn't work nor in CAS neither in Home.
however, please, see also this thread (for pivots calculation), that could help us to make an actual program to find Gauss elimination procedure, then pivots to use to find eventually LDLt decomposition.
Please, help EDIT:
working in progress in this thread in the Forum!
Reference URL's
• HP Forums: https://www.hpmuseum.org/forum/index.php
• :