Talk:Reduced row echelon form
"SWAP Rows" bug
The algorithm has a Bug.
It does NOT do the SWAP. No conditions are checked
It does NOT work for this example matrix
Solve the following system of equations.
3x+y− 4z=−1
x +10z= 5
4x+y+ 6z= 1
Solution. The corresponding augmented matrix is
3 1 −4 −1
1 0 10 5
4 1 6 1
Create the first leading one by interchanging rows 1 and 2
1 0 10 5
3 1 −4 −1
4 1 6 1
Now subtract 3 times row 1 from row 2, and subtract 4 times row 1 from row 3. The result is
1 0 10 5
0 1 −34 −16
0 1 −34 −19
Now subtract row 2 from row 3 to obtain
1 0 10 5
0 1 −34 −16
0 0 0 −3
This means that the following reduced system of equations
x +10z= 5
y−34z=−16
0= −3
is equivalent to the original system. In other words, the two have the same solutions. But this last
system clearly has no solution (the last equation requires that x, y and z satisfy 0x+0y+0z = −3,
and no such numbers exist). Hence the original system has no solution.
Umariani
- The algorithm does not have a bug. Failure to check conditions is not part of the algorithm, it is part of the implementation. That is just a degenerate case.
- A few points:
- The task does not require checking for degenerate cases.
- Even if it did, you gave no indication of which implementation fails on degenerate cases.
- What should the implementation do in the case of a "failure"? Error message? Warning? Or just precede as far as possible? (Which is what the Raku implementation does.)
- By the way, please sign your discussion page edits. --Thundergnat (talk) 12:38, 20 July 2023 (UTC)
- The algorithm is useful for solving a system of linear equations, but a reduced row echelon form can be made even if there is no such solution! The reduced form of the system
3 1 -4 -1 1 0 10 5 4 1 6 1 is 1 0 10 0 0 1 -34 0 0 0 0 1
--Wherrera (talk) 20:14, 20 July 2023 (UTC)
"Break" vs. "return" bug
The original author of the Python example mistakenly translated the keyword stop
that appears in the Wikipedia pseudocode as break
rather than the correct return
. This created a control-flow bug that didn't manifest itself when the program was run on the example matrix given in the task description, but did cause an exception if the program was run on, e.g.,
1 2 3 4 3 1 2 4 6 2 6 2 3 6 18 9 9 -6 4 8 12 10 12 4 5 10 24 11 15 -4
I noticed and fixed the bug a couple of days ago, but it seems that several of the other examples written before then (being, by and large, translations from the Python) copied the bug. Hence, I've marked all the examples that looked like they might have this bug with the needs-review template. Note that I erred on the side of false positives. That is, I'm pretty sure the examples I didn't mark are bug-free, but some of the ones I did mark may be fine. —Underscore 00:25, 5 May 2009 (UTC)
I ran the ALGOL 68 version and got:
(( 1.0000, 2.0000, 0.0000, 0.0000, 3.0000, 4.0000), ( 0.0000, 0.0000, 1.0000, 0.0000, 0.0000, -1.0000), ( 0.0000, 0.0000, 0.0000, 1.0000, 0.0000, 0.0000), ( 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000), ( 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000))
Vs the current python version which got:
1, 2, 0, 0, 3, 4 0, 0, 1, 0, 0, -1 0, 0, 0, 1, 0, 0 0, 0, 0, 0, 0, 0 0, 0, 0, 0, 0, 0
It looks like - currently- they are both getting the same answer... next step is it to get out a pencil.
NevilleDNZ 01:46, 5 May 2009 (UTC)
Bug in Common Lisp code
There is a bug in find-pivot that causes an infinite loop on some input. Here is a corrected version along with the matrix that results in an infinite loop with the current code. I tested the code with SBCL 1.0.40.0.debian on Ubuntu 10.10 64bit.
<lang lisp>(defun convert-to-row-echelon-form (matrix)
(let* ((dimensions (array-dimensions matrix))
(row-count (first dimensions)) (column-count (second dimensions)) (lead 0))
(labels ((find-pivot (start lead)
(let ((i start)) (loop :while (zerop (aref matrix i lead)) :do (progn (incf i) (when (= i row-count) (setf i start) (incf lead) (when (= lead column-count) (return-from convert-to-row-echelon-form matrix)))) :finally (return (values i lead))))) (swap-rows (r1 r2) (loop :for c :upfrom 0 :below column-count :do (rotatef (aref matrix r1 c) (aref matrix r2 c)))) (divide-row (r value) (loop :for c :upfrom 0 :below column-count :do (setf (aref matrix r c) (/ (aref matrix r c) value)))))
(loop
:for r :upfrom 0 :below row-count :when (<= column-count lead) :do (return matrix) :do (multiple-value-bind (i nlead) (find-pivot r lead) (setf lead nlead) (swap-rows i r) (divide-row r (aref matrix r lead)) (loop :for i :upfrom 0 :below row-count :when (/= i r) :do (let ((scale (aref matrix i lead))) (loop :for c :upfrom 0 :below column-count :do (decf (aref matrix i c) (* scale (aref matrix r c)))))) (incf lead)) :finally (return matrix)))))
(defvar *M* '(( 1 0 0 0 0 0 1 0 0 0 0 -1 0 0 0 0 0 0)
( 1 0 0 0 0 0 0 1 0 0 0 0 -1 0 0 0 0 0) ( 1 0 0 0 0 0 0 0 1 0 0 0 0 -1 0 0 0 0) ( 0 1 0 0 0 0 1 0 0 0 0 0 0 0 -1 0 0 0) ( 0 1 0 0 0 0 0 0 1 0 0 -1 0 0 0 0 0 0) ( 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 -1 0) ( 0 0 1 0 0 0 1 0 0 0 0 0 -1 0 0 0 0 0) ( 0 0 1 0 0 0 0 0 0 1 0 0 0 0 -1 0 0 0) ( 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 -1 0 0) ( 0 0 0 1 0 0 0 0 0 1 0 0 -1 0 0 0 0 0) ( 0 0 0 0 1 0 0 1 0 0 0 0 0 -1 0 0 0 0) ( 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 -1 0) ( 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 -1 0 0) ( 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0) ( 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0) ( 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1) ( 0 0 0 0 0 1 0 0 0 0 1 0 0 0 -1 0 0 0)))
(defvar *M-array* (make-array (list (length *M*) (length (first *M*))) :initial-contents *M*))
(print (convert-to-row-echelon-form *M-array*)) </lang>
Carlo Hamalainen 3:30PM, 6 Dec 2010 (GMT+10)
Fixed on main page.
Carlo Hamalainen 9:39AM, 16 Mar 2011 (GMT+10)
Code doesn't work for singular matrices
I tried putting in a singular matrix into the C# code and it is giving divide by zero errors. When I told it to skip when it is dividing by zero, the results keep changing. Also I corrected an error in the C# code that caused the lead value to be out of bounds.
- I won't be able to test it myself, but can you provide a sample matrix that reproduces the problem? --Michael Mol 03:08, 10 October 2010 (UTC)
- this gives a divide by zero:
- 1,0,1,0,1,0
- 1,0,1,0,0,1
- 1,0,0,1,1,0
- 1,0,0,1,0,1
- 0,1,0,1,1,0
- 0,1,0,1,0,1
- 0,1,1,0,1,0
- 0,1,1,0,0,1
- I haven't looked into it in detail, but adding an "if (div != 0)" line before the offending division seems to make it behave reasonably.
- edit: looking into it, the lead-- line seems to deviate from the original algorithm, replacing that with a "return matrix;" seems to fix it.
New test cases
Recently, I found that the earlier Java version did not work as expected due to floating point errors.
I recommend including these in your respecting languages: <lang> double matrix_2 [][] = { {2, 0, -1, 0, 0}, {1, 0, 0, -1, 0}, {3, 0, 0, -2, -1}, {0, 1, 0, 0, -2}, {0, 1, -1, 0, 0} };
solution: [[1, 0, 0, 0, -1] [0, 1, 0, 0, -2] [0, 0, 1, 0, -2] [0, 0, 0, 1, -1] [0, 0, 0, 0, 0]]
double matrix_3 [][] = { {1, 2, 3, 4, 3, 1}, {2, 4, 6, 2, 6, 2}, {3, 6, 18, 9, 9, -6}, {4, 8, 12, 10, 12, 4}, {5, 10, 24, 11, 15, -4} };
solution: [[1, 2, 0, 0, 3, 4] [0, 0, 1, 0, 0, -1] [0, 0, 0, 1, 0, 0] [0, 0, 0, 0, 0, 0] [0, 0, 0, 0, 0, 0]]
double matrix_4 [][] = { {0, 1}, {1, 2}, {0,5} };
solution: [[1, 0] [0, 1] [0, 0]]</lang>
- It's been ages since I have thought about this task. If we get differing results, how do we determine which result is "better". --Rdm 16:35, 16 January 2012 (UTC)
- I think matrix three is designed to show up floating point errors. Using the optional left argument for the
gauss_jordan
verb used in the J solution hints at that and using extended precision gives the desired answer. --Tikkanz 15:31, 26 January 2012 (UTC)
- I think matrix three is designed to show up floating point errors. Using the optional left argument for the
<lang j> 1e_14 gauss_jordan mat_3 1 2 0 0 3 4 0 0 1 0 0 _1 0 0 0 1 0 1.4803e_16 0 0 0 0 0 0 0 0 0 0 0 0
gauss_jordan x: mat_3
1 2 0 0 3 4 0 0 1 0 0 _1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 </lang>
Still a bug in java version?
I tried to run the the java code on matrix
1, 2, 3, 4, 3, 1 2, 4, 6, 2, 6, 2 3, 6,18, 9, 9,-6 4, 8,12,10,12, 4 5,10,24,11,15,-4
which correctly resulted in
1, 2, 0, 0, 3, 4 0, 0, 1, 0, 0,-1 0, 0, 0, 1, 0, 0 0, 0, 0, 0, 0, 0 0, 0, 0, 0, 0, 0
and then applied the code on this resulting matrix which generated an error. It seems that step 1 and step 2 do not recognize the correct pivot column for examples where there exist free variables in front of pivot columns (In the example above the second column of the resulting matrix). I corrected this by changing method isColumnZeroes not to consider the whole column: <lang java> public boolean isColumnZeroes(Coordinate a) { for (int i = a.row; i < numRows; i++) { if (matrix.get(i).get(a.col).doubleValue() != 0.0) { return false; } }
return true; } </lang> Now it works. All the other examples are also solved correctly. I have to admit that I changed the code so I am not sure if this error also occurs in the original version.
Edit: The new code is also not working, for instance matrix
0, 0 1, 1 -1, 0 0,-1 0, 0 0, 0 0, 0 0, 0 1, 1
results in
1, 1 0, 0 0, 1 0,-1 0, 0 0, 0 0, 0 0, 0 0, 0
In contrast, the code from http://ic.ucsc.edu/~ptantalo/math21/Winter07/GaussJordan.java works fine even though it does not implement the pivot stuff. Moreover, this code is understandable since it is not using crap functions like "break" and "continue" in loops which make code unreadable and error prone.
Bug in maxima code
For example, it fails with the following matrix:
1 1 1 1 1 0 1 1 1 1 0 0 0 0 1
rref produces:
1 0 0 0 0 0 1 1 1 1 0 0 0 0 1
The leading coefficient in the (3, 5)-th position is not the only nonzero element in its column.
Critique of... something?
Hello RosettaCode ...
Your algorithm/function does NOT work ...
The "SWAP" is not being done. ...
Try this matrix to see the problem. ...
//===================================
Linear Algebra with Applications by W. Keith Nicholson, University of Calgary
page 12 - Systems of Linear Equations
Solve the following system of equations.
3x + y − 4z = −1
x + 10z = 5
4x + y + 6z = 1
Solution. The corresponding augmented matrix is
3 1 −4 −1
1 0 10 5
4 1 6 1
Create the first leading one by interchanging rows 1 and 2
1 0 01 5
3 1 −4 −1
4 1 6 1
Now subtract 3 times row 1 from row 2, and subtract 4 times row 1 from row 3. The result is
1 0 10 5
0 1 −34 −16
0 1 −34 −19
Now subtract row 2 from row 3 to obtain
1 0 10 5
0 1 −34 −16
0 0 0 −3
This means that the following reduced system of equations
x + 10z = 5
y − 34z = −16
0 +0 +0 = −3
is equivalent to the original system. In other words, the two have the same solutions.
But this last system clearly has no solution (the last equation requires that x, y and z satisfy 0x+0y+0z = −3,
and no such numbers exist).
Hence the original system has no solution.Big text
//===========================
User:Umariani (Moved from main page by --Thundergnat (talk) 11:06, 20 July 2023 (UTC))