#4562 guess_exact_value failure on trivial case

[Stavros Macrakis]

guess_exact_value(float(2*sqrt(2))) => 2.8284271247461907

guess_exact_value can't always give a symbolic result, but it fails in too many trivial cases: I'd have thought that n*sqrt(m), n+sqrt(m), sqrt(m)+sqrt(n), etc. would almost always succeed for small m, n. But in fact they often fail. Is that because many cases are "close calls", because the algorithm isn't very powerful, or because our parameter settings aren't very good?

Maxima 5.47.0 SBCL 2.4.9

for i:2 thru 5 do for j:i+1 thru 10 do if floatnump(guess_exact_value(ff:float(sqrt(i)+sqrt(j)))) then print([i,j,ff]);
[2, 4, 3.414213562373095] 
[2, 6, 3.863703305156273] 
[2, 7, 4.059964873437686] 
[2, 9, 4.414213562373095] 
[2, 10, 4.576491222541475] 
[3, 4, 3.732050807568877] 
[3, 6, 4.1815405503520555] 
[3, 7, 4.377802118633468] 
[3, 8, 4.560477932315068] 
[3, 9, 4.732050807568877] 
[3, 10, 4.894328467737257] 
[4, 5, 4.23606797749979] 
[4, 7, 4.645751311064591] 
[4, 8, 4.828427124746191] 
[4, 10, 5.16227766016838] 
[5, 6, 4.685557720282968] 
[5, 7, 4.8818192885643805] 
[5, 8, 5.0644951022459805] 
[5, 9, 5.23606797749979] 
[5, 10, 5.39834563766817] 


for i:2 thru 10 do for j:i+1 thru 10 do if floatnump(guess_exact_value(ff:float(i+sqrt(j)))) then print([i,j,ff]);
[2, 3, 3.732050807568877] 
[2, 5, 4.23606797749979] 
[2, 7, 4.645751311064591] 
[2, 8, 4.828427124746191] 
[2, 10, 5.16227766016838] 
[3, 5, 5.23606797749979] 
[3, 6, 5.449489742783178] 
[3, 7, 5.645751311064591] 
[3, 8, 5.828427124746191] 
[3, 10, 6.16227766016838] 
[4, 5, 6.23606797749979] 
[4, 6, 6.449489742783178] 
[4, 8, 6.828427124746191] 
[4, 10, 7.16227766016838] 
[5, 6, 7.449489742783178] 
[5, 7, 7.645751311064591] 
[5, 8, 7.828427124746191] 
[5, 10, 8.16227766016838] 
[6, 7, 8.64575131106459] 
[6, 8, 8.828427124746192] 
[6, 10, 9.16227766016838] 
[7, 8, 9.828427124746192] 
[7, 10, 10.16227766016838] 
[8, 10, 11.16227766016838] 
[9, 10, 12.16227766016838] 

[Robert Dodier]

Stavros, I don't see that output; the loop doesn't print anything in either case, and trying similar expressions seems to show guess_exact_value returns reasonable results (not exactly the same as the input, maybe factored differently or, in some cases, with nested sqrts).

I tried it with 5.47.0 and with a recent build, on Linux in both cases. What does build_info() say on your system?

What do you get if you say batch("rtest_pslq", test) ? That has a number of simple examples for guess_exact_value. I get no errors; do any of the examples work when you try it?

[David Billinghurst]

I just tried on cygwin/clisp/windows11 with 5.47.0 and git head. Same results as Robert.

(%i2) build_info();
(%o2)
Maxima-version: "5.47.0"
Maxima build date: "2023-06-25 17:16:51"
Host type: "x86_64-pc-cygwin"
Lisp implementation type: "CLISP"
Lisp implementation version: "2.49+ (2010-07-17) (built on Walter [192.168.178.24])"
User dir: "/home/dabilling/.maxima"
Temp dir: "/tmp"
Object dir: "/home/dabilling/.maxima/binary/5_47_0/clisp/2_49___2010_07_17___built_on_Walter__192_168_178_24__"
Frontend: false

[Robert Dodier]

I'm closing this as not-a-bug since we weren't able to reproduce it. We can reopen it if someone figures out how to reproduce it.

[Stavros Macrakis]

Sorry, I had apparently changed fpprec and forgotten that it affects guess_exact_value.