Error found on HP49G
#1

I used the NUM.SLV on th HP49G to solve linear sys of the following :

6x-3y=15 -2x+y=0

there is no solution , but the HP49G find the solution to be x=1.8 and y=-0.9, I tried other non solution linear system and still the HP49G insist to "solve" it and report as it have solution for example I tried also the following: x-y=2 2x-2y=0 the HP49G find the solution to be x=0.2 and y=-0.2. those any one can tall me what is wrong or it is a BUG!!

#2

Dear Aharon: I do not have a 49G calculator, but I put your equation 6X-3Y=15 2X+Y=0 into a 48GX solve for linear equations and came up with the answer that you did X=1.8 and Y=0.9. Here is your explanation. First off in general, the solution set for a system of two equations in two variables is one ordered pair, an infinite number of ordered pairs or an empty set. This equation solution is an infinite number of ordered pairs, and the 48 and 49 will give you this answer if you investigate further. Right off the bat, you should enter the equations in matrix form on the stack, press MTH MATR NORM COND (hp 48G,GX) and compute its condition number. If it is large, 10 to the 12th power or higher, then it is an ill conditioned matrix. In this case with the above equation, it returns a condition error infinite results. Meaning, there are infinitely many ordered pairs. I'm sure the hp49G has the same tests. With this test, you can determine before hand if the equation is solvable with one specific ordered pair. If you do solve it first without testing, there is also another way to test. Your hp 49 returned an answer to you that was not useable because it returned the smallest ordered pair that it could obtain, basically it returned one of the infinite pairs as an answer. Finding the residual of the array will verify this. I will not go into these instructions as I do not have a 49G but I'm sure it has a test for the residual of an array or matrix. The smaller the absolute value of the elements the better the solution, in this case on my 48GX it was rather high. So it is up to the individual to test the answer they receive from the calculator for validity. This is true in both the 48&49. In the hp48GX manual it says that the Linear system solver will return a result array for any of the following systems. Exactly determined systems, Over determined sytems, or under determined sytems, and goes on to say to pay attention to your linear system because it will influence the way you should interpret the result array, and to test for ill conditioning before you solve, or even accept an exact solution as a valid solution. No bug! Hope I have helped you.

#3


> 6x-3y=15 -2x+y=0

On my HP41CV the SIMEQ program of the Math I rom module displays: NO SOLUTION. The answer of my HP15C is instead [-7.5e10, -1.5e11].

> x-y=2 2x-2y=0

Again, the HP41CV + Math I rom module displays: NO SOLUTION. The answer of the HP15C is [2e10, 2e10].

The HP15C Advanced Functions Handbook explains the problem of errors in matrix calculations: a square matrix A is said to be ill-conditioned (nearly singular) if the condition number K(A)=|A|*|A^-1| is very large. ( where |A|=row norm of A, |A^-1|=row norm of the inverse of A ). A large condition number makes possible a large error in the result.

This seems to be the case of both your examples: K(A)=1.2e11 and K(A)=6e10 respectively (HP15C results).

The handbook discusses also the accuracy of numerical solutions to linear systems, describes how to improve the solution precision by iterative refinement (residual correction), round-off errors in internal calculations, accuracy of numerical calculations, and so forth. Similar handbooks are available for the HP28 and the HP48, if I'm not wrong.

The main problem with the HP49G is that it isn't a real HP engineering calculator. :-)

#4

Thanks for your answer. But I think your HP15C solution is not relevant here, since the solution you got is different from mine, (very different !!) any way on "non solution equations" I expect my HP to report me that it is a "non solution equations" and not give me a solution that it is not mathematically right. on 2 order equation it's simple to check out the HP but what about 19 order equation that I need to use??.

#5

Thanks for your answer. I disagree whit the idea that this 2 order equation has infinity solutions, that is way:

6x-3y=15 -2x+y=0

that mean y=2x

so 6X-3(2x)=15

and 6x-6x=15

finely 0=15 and that can not be!!! no matter what is the reason the HP must report that this 2 order system has no solution !!! I usually need to calculate 13-19 order system and such results must be reported !!

#6

I tried this system on my TI-92+ and it correctly reported "No solution".

#7

> Thanks for your answer. But I think your HP15C solution > is not relevant here, since the solution

It looks like there is a little misunderstanding. The point is: in order to make proper use of numerical computation algorithms it is required to learn how they work.

#8

Just for fun, I decided to try this on my HP-41CX. Using the MATH I ROM, I got the same results as your 41CV for both sets of equations: NO SOLUTION. However, using the MSYS function from the HP-41 Advantage ROM, I got [2.5e10, 5e10] for the first pair and the same as your 15C for the second pair [2e10, 2e10]. After throwing together a quick program to calculate K(A) using RNRM and MINV, I got 1.2e11 for the first matrix and 6e10 for the second (both the same as your 15C's results). Surprisingly, I didn't find any mention in the documentation of checking the condition number or of the possibility of errors in the solution matrix. I guess they expected that anyone using these functions would be aware of these things, but for math non-experts (like me!) it would have been nice to mention them.

#9

One more note on this issue... I have a HP49G and obviously get the same (incorrect)result. My stepson's TI89, one of the HP49G's biggest competitors, gives the correct solution. I just have to shake my head on this one.

#10

The determinant (as calculated on a 48G) is zero. The condition value is infinate. What do you expect?

#11

Depends how you look at it. The numerical algorithm to solve a set of linear equations is the same in the 48 and 49. (In exact mode, the 49G does return [0 '?'] to indicate a defective set). The algorithm is designed to perturb the original matrix with small quantities to reach a solution if no solution exists. While I contend this practice, it is true that an exact linear dependency will only be discovered for the smallest of problems (talking about numerical, approximate mode here), leading you to believe that if a solution has been found, it must necessarily be the correct one. Not so.. rule of thumb: always check the condition number.. its exponent tells you how many digits you can trust of the answer ie if you have a condition number of 1e5, you can trust 15-5=10 digits of the answer (for exact input, otherwise it's 12-5=7 digits). If your condition number exceeds 1e15 (or 1e12) the answer is useless, and no numerical algorithm can help that. It would have been nice if a command existed that solved the system AND returned the condition number at the same time..

#12

I disagree whit the idea that this 2 order equation has infinity solutions, that is way:

6x-3y=15 -2x+y=0

that mean y=2x

so 6X-3(2x)=15

and 6x-6x=15

finely 0=15 and that can not be!!! no matter what is the reason the HP must report that this 2 order system has no solution !!! I usually need to calculate 13-19 order system and such results must be reported !!

#13

can you be aware of these things when you using 19order system??????

#14

Dear Aharon: I totally agree. The answer to the problem 6X-3Y=15 -2X+Y=0 is a disjoint set. But I guess I was not clear enough in my answer that the HP 48GX returns an infinite results answer when testing the condition of the matrix. I have had this question before, and the hp calculator support division has told me that the 48G series will sometimes return a no solution answer, but not always, as the software will sometimes change a system slightly to arrive at a "close" answer. But the hp personnel say this was the way it was designed. I do not agree with this. You can view the graph of this sytem with conic plots form and enter both equations in a list with delimiters and it will plot parallel lines. This is a disjoint set. I think the 48&49 in this case is only plotting the solution to the first equation, giving all points on a line as the solution, maybe, giving you a misleading answer, but that is why you can test the residual. I only take an answer as the gospel if the residual returns a set [0 0]. But as far as reporting in all cases exactly what is being returned in a system will not happen. That is why if I am not sure, I will graph the system and look at it. The Texas Instruments graphing calculators error message are more informative, and if you need to always know how a system is behaving without further testing, maybe you should trade yor hp49 for a TI89. I'll probably have my tongue cut out for that one. One other suggestion. If you are doing many large systems, you should probably be doing these on a computer. Happy calculating! Bill.

#15

Thanks, I understand what you mean . and about TI 89/92 I am going to seriously think about it....

#16

Have you contacted HP, via their web site, about this problem?

From the other responses from more learned HP users it may well be possible to derive the accuracy (none in this case) of the result. However using the HP49G guides (which is all you can assume) there is no mention of further investigation to be carried out. The ordinary user will probably trust that the answer is right.

As with some other areas of the HP49G's performance it is too variable, this is the worst possible situation as any non-expert user can no longer trust the answers given.

It is no good to say 'well the user is inputting impossible problems and the HP49 is just doing the best it can'.

If it is true that the TI89/92 (and even the HP41!) gives the more informative and 'correct' result then this is bad news for the sales of HP49G and needs to be rectified.



Possibly Related Threads…
Thread Author Replies Views Last Post
  MS advert shows spreadsheet with obvious error BruceH 3 2,146 11-14-2013, 09:50 AM
Last Post: Bill (Smithville, NJ)
  [HP Prime] Calculating Prandtl Number with Units (bug found in USIMPLIFY) Timothy Roche 1 1,326 11-13-2013, 04:07 PM
Last Post: cyrille de Brébisson
  OT--Found!!! An HP reference on 'The Big Bang Theory' Matt Agajanian 0 1,094 11-08-2013, 10:23 PM
Last Post: Matt Agajanian
  HP Prime: Rounding error in determinant Stephan Matthys 3 1,607 10-25-2013, 09:29 PM
Last Post: Walter B
  Prime Error or Mine? toml_12953 12 3,532 10-22-2013, 10:35 AM
Last Post: toml_12953
  Explaination on How to Reset Caculator in Users guie: error Harold A Climer 5 2,514 10-15-2013, 02:11 AM
Last Post: cyrille de Brébisson
  Repair of HP-34C - Error 0 and Error 9 Jeff Kearns 3 1,644 10-11-2013, 12:29 PM
Last Post: Randy
  Do You Think a Listing Error Was Made? Jim Johnson 13 3,776 09-04-2013, 09:23 AM
Last Post: Eddie W. Shore
  HP-fonts - Where can I found? Csaba Tizedes (Hungary) 4 1,610 08-23-2013, 05:01 PM
Last Post: Csaba Tizedes (Hungary)
  HP Prime: Error reports : Here Patrice 111 22,324 07-24-2013, 05:52 PM
Last Post: Thomas Klemm

Forum Jump: