docsox,
Your caution is reflective of experience and a lesson for everyone. Your post made me think about this a little more and come to believe more firmly that there is very little that is new in Log Point's basic technology. In fact, it makes me curious as to what exactly they have patented, but more on that later.
First of all, I am not a mathematician. I am a software engineer. So I can not speak with any more expertise than that. However, I have implemented floating point math routines in software and made use of them for basic math functions. So I have some practical experience.
Generally, floating point numbers can be represented in a number of ways in the software world of 1s and 0s. In order that most applications understand each other the engineering organization IEEE ("I, triple E") has established the standard 754 representation. As it turns out, the software I wrote doesn't use this standard, but it doesn't matter. One simply needs to be able to add, subtract, multiply, divide etc. the numbers, however you represent them. To do this, you basically implement the functions similarly to how you do it by hand. So, notice how easy it is to do 3.14116+4.878. It's not as easy as 3+4 but not too hard either. But look at how many steps are involved in 3.141x4.878 or 3.141/4.878. Those are a little more nasty - in software too! Wouldn't it be nice to make the multiply and divide as easy as the addition! And you can,if you were to represent these numbers by their logarithms.
What is a logarithm? It is a number that represents the power to raise some base number (often 10) in order to get the desired number. In other words, if I know that 10 ^ 2 (10 to the second power) = 100 then I know that log(100) = 2. A well known, and helpful, mathematical principle is that one can do multiplication and division quite easily by adding or subtracting the logarithms of numbers. In other words n=3x4 can be done by doing log(n)=log(3)+log(4) and then converting the result back to a non logarithm by doing 10^n. Let's see: 3x4=12. log(3) = .477. log(4)=.602. .477+.602=1.079. 10^1.079=12! Wow! Believe it or not, this is the principle upon which SLIDE RULES were based. That's why people used them!
Ok, but it's still a hassle! Right. But then it's a MAJOR hassle(and slow)to do 3.141x4.878 in ones and zeros too! But once people implemented it in calculators, everyone tossed their slide rules and thought life was good! But, man, once we're really serious about crunching millions of floating point numbers it's stinking slow! So, let's say we make it easy to convert to and from logarithms somehow, say some technique with lookup tables (hint: see LGPT's home page). Better yet, let's just say that we break convention and define a way of continuously representing floating point numbers in their logarithm form. So I might type "3" on my spread sheet but the computer always thinks of it as .477. (Hint 2: See LGPT's home page). That's different, but ok. Now, say I want 3.141x4.878. Hey, now that's easy, just .497+.688 = 1.185. Well, I need to convert it from a logarithm => 15.322. Yup.
Assuming the above is essentially correct the following observations/questions arise:
1. Logpoint's "technology" is in a manner of speaking, a new application of "slide rule" technology.
2. The "technology" isn't quantum mechanics (i.e. no worry that the technology is "smoke and mirrors"), but it's a clever application of rather basic math - In other words, no one appears to have thought of doing computer floating point this way before.
3. It makes me wonder what parts of this they actually have patented. The conversion process, the applications? You can't patent logarithms.
I think we need to ask the company about this as well as look up the patents. I think sometime in the near future I can go down to the university and do a patent search for the patents listed on their home page. It may not be this week, however.
By the way, everything here is the insight and the application of the insight. In other words, the value in this company isn't in its complex math. I, a mere 31 yr old engineer, could write software to do most of this. The value has to be in its patents and in proving the value of the techniques to other companies who need to floating point math, and in establishing a new standard on the basis of the previous two items. The conversion of these techniques to hardware, as they are doing, is, really, a no risk - just takes some $. From what LGPT has stated, they seem to have lots of interest. So everything, it seems to me, hinges on the patents.
IMO,
Mark N.
|
