20100106, 17:27  #1 
∂^{2}ω=0
Sep 2002
República de California
11740_{10} Posts 
The Future of Computational Mathematics
While searching for something on a related topic on the web, I came across this interesting 2005 paper from Jonathan Bailey and Peter Borwein (it's a pdf), summarizing the state of the art computational mathematics:
The Future of Computer Assisted Mathematics 
20100107, 04:53  #2 
"Richard B. Woods"
Aug 2002
Wisconsin USA
2^{2}·3·641 Posts 
I'm reminded of the history of chessplaying computers.
In the 1970s, they were pretty bad  I, a class B player at roughly the 65th percentile of USCF players, could easily and consistently beat the ones marketed in games shops. Then there were steady improvements, so that the top chessplaying computers (either just software running on ordinary generalpurpose hardware, or some cases of specialpurpose hardware optimized for investigating future chess moves) ranked as class A, then Expert, then Masterlevel. It became necessary to add tournament rules governing how chessplaying computers could participate as players in tournaments. Tournament organizers were given the right to exclude all computer participation if they wished. Top computers began to challenge (or, rather, their human handlers issued challenges on their behalf) grandmasters, to the extent of winning a significant proportion of games played in international tournament conditions. Then they challenged the world champion and made a decent score in matches. Then they beat the human world champion  by a margin similar to that by which the human champion had beaten his human predecessors. In the early years, it had been thought that the most effective way to improve chess computers was to employ grandmasters to codify their knowledge and techniques into the software. That is, making computers play more and more like humans was thought to be the road to success. The application of more and more sheer brute speed in looking ahead at the possibilities in a game seemed to be ineffective. But then the brutespeed method began to surpass the expertknowledge method. The strongest chess computers began to be the ones that could look ahead and evaluate the largest number (plies) of future move candidates in the game time available for each move. It now appears that a certain number of brutespeed plies is sufficient to surpass the lookahead capabilities and intuition of the best human players. Computers now sometimes win games with eerie move combinations that no human player could imagine discovering, because there is something about those combinations that was just too strange for humans to conceive  even when the depth of lookahead is the same as for combinations more readily found by grandmasters. It seems that chess computers have revealed that there are certain types of chess move combinations to which humans are "blind". The fact that this was revealed by the brutespeed method, rather than the expertknowledge method, is disturbing to many. This toostrangeforhumanstoimagine phenomenon may occur in computational mathematics, too. Once the math computers have advanced sufficiently, they may find things that will seem too strange for humans to imagine ever discovering by themselves, even though the computers discover these things via a brutespeed approach. (I don't mean, for example, finding that the fourcolor theorem is proven by considering thousands of cases. That, I think, was readily conceivable by humans, even if they thought it unlikely. I mean stuff that is too strange for anyone to have imagined beforehand.) Last fiddled with by cheesehead on 20100107 at 04:54 
20100107, 06:32  #3 
Jul 2003
So Cal
2·3·11·37 Posts 

20100107, 06:33  #4  
Dec 2008
7^{2}·17 Posts 
Quote:
Last fiddled with by flouran on 20100107 at 06:50 

20100107, 13:51  #5 
(loop (#_fork))
Feb 2006
Cambridge, England
14466_{8} Posts 
I am probably too ignorant of thetafunctions, hypergeometric sequences, polylogarithms, modular equations and the works of Ramanujan validly to make this statement, but I don't think the expressions in the paper would have shocked Ramanujan.

20100107, 13:58  #6 
Aug 2006
5,981 Posts 
Yes, the statement should be amended to "computers may find things that will seem too strange for humans (other than Ramanujan) to imagine".

20100107, 20:54  #7 
Feb 2005
11×23 Posts 
A relevant essay dated the same year 2005:
Brian Davies Whither Mathematics? And yet another paper on the topic: Towards Computer Aided Mathematics . Last fiddled with by maxal on 20100107 at 20:59 
20100107, 22:31  #8 
Jan 2008
France
238_{16} Posts 
Borwein and Bailey published some books about experimental math: http://www.experimentalmath.info/books/
I wonder if they are good or not. 
20100107, 22:56  #9 
Mar 2004
17D_{16} Posts 
about eing careful about approximations...
I remember that i have read somewhere that there is an infinite sum which converges to some value very close but not pi. (There were similar examples in this paper). The sum is accurate to some billion digits, but after that is wrong. Does anybody know which equation is meant? I don't remember that equation or the place i have read it. 
20100107, 23:23  #10 
(loop (#_fork))
Feb 2006
Cambridge, England
2×7×461 Posts 
Sums of exp(a*x^2) get accidentally very close to sqrt(pi), for reasons related to the fact that the Fourier transform of exp(a*x^2) is the very fastshrinking (if a is small) function exp(1/a * x^2):
so sum(t=400,400,exp(t^2/100)) is within 10^420 of 10*sqrt(Pi) I think this may be what you're thinking of Last fiddled with by fivemack on 20100107 at 23:23 
20100108, 03:44  #11 
"Lucan"
Dec 2006
England
2·3·13·83 Posts 

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
a computational perspective and observational N.T.  MattcAnderson  Miscellaneous Math  0  20170630 11:33 
Mathematics Diary  Nick  Math  0  20170502 09:58 
Computational predictions!  Xyzzy  Lounge  20  20151010 02:07 
Mathematics Survey!  mfgoode  Information & Answers  0  20070721 15:38 
mprime: is this a computational error or ?  gnychis  Software  3  20060207 01:47 