|Re: Many thanks Valentin [LONG]|
Message #37 Posted by Vincent Weber on 19 Nov 2003, 11:47 p.m.,
in response to message #29 by Valentin Albillo
Thanks to you for kindly accepting my apologies, the honor is mine indeed. Now, let's quickly forget this most unpleasant affair and move on :-)"
My pleasure. Yes, let's move on ! :-)
"That's *not* the case with the real 15C, where such operations as 3^201 is off by *two* counts in the last place (worst case known. See "Advanced Functions" manual)".
Unfortunately, I do not own a real 15C.I have been searching for one for month in Singapore... I managed to get one 41CV, one 41CX, one 42S and one 32SII. 15C, niet - probably the owners are too attached to it to sell it... ;) I don't own the "Advanced functions" either - it does not seem to be available on the Internet...
So, we will complement each other: I will tell you the behaviour of Pocket15C, and you will be able to compare it with the real behaviour :)
3x201 answers 7.968419666E95 on Pocket15C, and 7.96841966626E95 on my 48GX. It seems that the first 10 digits are correct then ?
"Does the display blink continuously when some operation results in overflow ? If yes, does "backarrow" stop the blinking ? Does CF 9 stop the blinking ? Does any other key or operation stop the blinking ? Does SF 9 start the blinking again ?"
The display does not blink: as explicitely documented by Lygea, the digits turns red instead. However, the behaviour seems to be the same than on the real 15C, i.e. both "backarraw" and CF 9 get the digit back to black, while SF9 get them back to red.
"When running a program, do "user STO" and "user RCL" automatically skip the next step when the last element is processed ?"
I have tried the following programs:
001 f LBL A
005 f DIM A
006 f MATRIX 1
007 f LBL 0
009 USER STO A
010 GTO 0
After inputing 9 elements, the program exists the "infinite" loops and displays "123" which I believe is the expected behaviour.
USER RCL (lauched immediately after 'USER STO')
001 f LBL A
002 f MATRIX 1
003 f LBL 0
004 f PSE
005 USER RCL A
006 GTO 0
Same bahaviour - the program successfully exists the loop.
"When running a program, does MATRIX 7 (Frobenius norm) automatically skip the next step if the contents of the X register is *not* a matrix descriptor ?"
I have tried the following programs (assuming A holds a 3x3 matrix as in the previous example):
001 f LBL A
002 RCL MATRIX A
003 f MATRIX 7
006 g RTN
001 f LBL A
003 f MATRIX 7
006 g RTN
Again it seems that the behaviour is correct.
"When running a program, does SOLVE automatically skip the next step if no root was found ? Does INTEGRATE skip the next step if the final value didn't meet the precision requirements after reaching the internal maximum iteration limit ?"
For SOLVE, I tried the following:
001 f LBL A
002 g X^2
005 g RTN
(f(x)=x^2+1, not likely to find *any* root unless the solver has been upgraded to a complex one ;))
006 f LBL B
007 f SOLVE A
The output is the expected "2". If I suppres the "g X^2" step in A, i.e. if I solve for X+1=0, I get "12" as an output for B.
I failed to reproduce this with a few examples. I am not too sure on how to get such a 'difficult' function to integrate for the 15C... Any idea ?
"Do all complex multi-valued functions have the same branch cuts as in the HP-15C ? Check square root, exp, log, and hyperbolics"
"1 ENTER f I sqrt g ->P" (in RAD modes) gives sqrt(sqrt(2)) as module and PI/8 for the principal root of 1+i = sqrt(2)*exp(i*PI/4). I would say that this is the expected behaviour ?
"1 PI 4 / f I EXP g ->P" (in RAD modes) gives the expected e as module, and PI/4 as argument.
"1 ENTER f I g LOG" gives the expected LOG(Sqrt(2)) as a real part, and PI/4/LN(10) as the imaginary part: LOG(1+i) = LOG(sqrt(2)*EXP(i*PI/4)) = LOG(sqrt(2)) + LN(EXP(i*PI/4))/LN(10) = LOG(sqrt(2) + PI/4/LN(10).
"1 ENTER f I f HYP SIN" gives the same results as with using the definition: (exp(x) - exp(-x))/2
"Do the results agree with the real 15C for inverses, determinants and system solutions for singular or very nearly singular matrices ? I would try all the examples in "Advanced Functions", just to be sure"
I can't comment on that, but I can try whatever examples you have. If I'm not wrong, that was done once on this newsgroup with correct results.
"Does the random number generator mimic the results of the one in the real 15C ? If not, does it pass the Spectral Test (as the real 15C does) ?"
I cannot comment about the results on the 15C. As for the spectral test, I heard of it, but I have no time right now to try it :)
"Does [Pi] [STO #RAN] [RCL #RAN] produce 0.3141592654 ? Same with 9.999999999E99, -9.999999999E99, 9.999999999E-99, -9.999999999E-99"
With PI, the result in 0.314159265; the others triggers an overflow (numbers become red), and the answer is either 1.0 or -1.0. Do we have a pitfall here ?
"Does line number branching work the same as in the real 15C ? I would check with out-of-range, negative, and non-integer values in I."
The bahaviour seems to be proper. Out-of-range values triggers an 'Error 4'.
"Does GTO I work correctly when I contains a matrix descriptor ?"
This does not trigger an error, but do be honest, I am not 100% sure of what this is supposed to do ?
"if you are in complex mode, can you store a matrix descriptor in both the real X-register *and* the imaginary X-register ? What does it do if you then press [+] or [-] ?"
RCL MATRIX A f I triggers an "Error 1"... I have tried that in the beginning hoping that this would have the same bahaviour as the 42S to create a complex matrix, and was disappointed when I saw that the only way to create complex matrix is to transform them from 'real' matrixes. Do you mean that this is actually supposed to work ? (forgive my ignorance - please keep in mind that I could not find a real 15C and that this simulator is all that I have !").
"Are both RCL DIM (i) and RCL DIM I legal instructions ? What do they do ?"
Both of them are legal. Assuming I contains a 3x3 matrix descriptor, RCL DIM I gives the dimensions of the matrix on the stack (3 and 3), while RCL DIM (i) gives 65, which I strongly suspect to be the number of registers used by the matrix + 1 (65 as a fixed-sized in the case of Pocket15C).
"I wrote a very good one as part of an HP-41C Solutions Book published locally (Matematica Avanzada). It's never been published elsewhere, but regrettably I don't have a copy of the book right now. I know I *do* have the original listings and documentation, but they're buried in a 4-feet stack of old listings and calculator docs and materials, and it would take some time to search for it, then retyping it all in a word processor. I'll do it, eventually, but it will take some time. I remember it accurately implemented an RPN complex stack just as a user would expect it, including all functions and functionalities, even complex store and recall, programmability, anything. It was a large, carefully crafted program and it certainly delivered the goods. I found it superior in usability to even the HP-15C's native complex mode, as it offered important extra functionality, such as complex storage/recall. It also had a novel and imaginative solution for conveniently implementing complex numbers entry and display"
I am looking forward to see it. Congrats ! :)
"... and nevertheless, it doesn't really matter, as most any contemporary SHARP models (and even some CASIOs) can run rings around them both ... ;-) No flames, please, just kidding ! :-)"
Actually, I have seen some many people bashing Casio here... while I do esteem Casio ! My first machine was a Casio fx-8000G in high-school, and was doing a good job. Casio has been innovating in may ways: in 1982, the fx-602p, while not as powerful as the HP41, was faster, with a real LCD screen; the FX-702P in 1981 was a faster competitor to the Sharp PC-1211; in 1985, Casio simply revolutionnized the market by creating the first graphing calculator (the FX-7000G) with a large screen, together with the first programmable formula-based calculator (the FX-4000P, which is an adorable little thing that I have in my collection). Speed and quality of screens were key assests to Casio in the 80's. HP28C/S came only 1/2 years later, with half of the size of the screen, and were very expansive. During my school days, the usual path was to have a Casio in high-school and to move to the prestigious new HP-48SX, which was actually my first beloved HP. It is only recently that I started to discover the virtues of older machines, such as the 41,15, and 42 :)
In comparison, TI did not innovate much: With the expection of the TI-59, a prestigious machine with modules (in 1977 !) which was litterally killed by the HP-41, TI was down until the '90s (I do not think of the TI-66 as a serious rival to the HPs, and the TI-88 project was cancelled). And the machines of the '90s were actually copies of Casio (think of the TI-81, so simular to the Casio FX-7000G... 5 years later !). Only their TI-92 could 'beat' the HP48, but I do not call this a calculator anymore, given the volume and weight :) They are finally dominating the market with the TI-89, a fine but overkill machine, which is actually very recent (1998).
So, long live Casio !
My 2 cents :)