On Tue, Feb 12, 2008 at 03:04:13AM -0800, Andrew Lentvorski wrote:
> >Ever heard of the chain rule?  df/dx = (df/du) * (du/dx) .
> >
> >Technically this is not correct because f is either is function of x
> >or u but not *both*.
>
> It is quite technically correct.  If you write out the whole partial
> derivative mess and then collapse the terms that go away because u is a
> function solely of x, your formula is quite fine (in fact, all of the
> calc books *I* have do indeed give a pointer to the fact that the chain
> rule is a simplified partial derivative even if it is in a footnote).

*We* both understand what the chain rule really means.  Notice it took
you a whole paragraph to decode your understanding.  You mentioned
simplifications, footnotes and other gymnastics.  Why make the poor student go
on a scavenger hunt for all that implicit knowledge?  Sussman is just
suggesting that you can focus all that gymnastics into a laser sharp computer
program.  That's all we're saying.  Nothing more.

> We model at varying levels of abstraction and call them correct.  This
> is quite acceptable, TYVM.  And, in this instance, the result is
> actually correct.  Be careful calling things "wrong" when they are not.

Yes but you need *precise* way to communicate even just within your chosen
layer.  I'm not saying abstractions and layers are a bad idea.  We're just
suggesting the software is superb method of precise *communication*.

> The ability to abstract at different levels of correctness and
> completeness is a hallmark of modern science and engineering.  The
> inability to accept different levels of abstraction and correctness is a
> hallmark of mathematics.  Both have their place--see: Dirac delta function.

Straw man.  See my comments above.  I'm not arguing against abstraction.

> And what have *you* been smoking.  Students learn what gets them through
> their immediate problem.  I'm not much different.

Yes and a software algorithm would be the most efficient means to get through
immediate problems.

> I'm not going to learn all of formal lambda calculus to understand
> lisp/scheme.  I'm not going to learn deep pure math to balance my
> checkbook.  etc.

Another straw man.   See above again.

> Sorry, normal programming languages are *lousy* constructs to match to
> mathematics and physics.

Maybe, maybe not.  But informal English verbage is even more lousy.

> Why do you think Mathematica and Matlab are so amazingly popular?

These would be fine too.

Chris

-- 
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-lpsg

Reply via email to