[SciPy-Dev] Incorrect results from scipy.integrate.romberg

[Pauli Virtanen]

Wed, 28 Jul 2010 17:49:55 -0400, Thomas Robitaille wrote:

> Hi,
> 
> I think there is some kind of bug with the romberg integration function.
> In the following code, romberg integration give an incorrect answer that
> is too large by a factor of two:
> 
> from scipy.integrate import quad, romberg
> 
> def powerlaw(x):
>     return x**3
>     
> print "%12.3e" % quad(powerlaw, 1e1, 1e5)[0] # gives 2.500.e+19 print
> "%12.3e" % romberg(powerlaw, 1e1, 1e5) # gives 5.000.e+19
> 
> The correct answer is 2.5e19, but romberg gives twice that.
> 
> Also, as in my previous email, it seems the tol= argument in romberg()
> is the absolute tolerance, not relative. Would it not be a good idea to
> be able to specify a relative tolerance?

Yes, fix'd.

Also patches (that include tests) are accepted in the future :)

-- 
Pauli Virtanen