[Rd] eigen()

Prof Brian Ripley ripley at stats.ox.ac.uk
Tue Jan 10 16:01:00 CET 2006


I haven't seen most of this thread, but this is a classic case of passing 
integers instead of doubles.  And indeed

     else if(is.numeric(x)) {
 	storage.mode(x) <- "double"

has been removed from eigen.R in R-devel in r36952.  So that's the 
culprit.

[BTW, x86 is not 64-bit, which is x86_64.  I'd take x86 to be ix86, what 
Intel calls ia32 (and AMD does call x86, given what 'i' stands for here).]

On Tue, 10 Jan 2006, Robin Hankin wrote:

>
>
> On 10 Jan 2006, at 14:14, Peter Dalgaard wrote:
>>
>
>
>>>> Strange and semi-random results on SuSE 9.3 as well:
>>>>
>>>>
>>>>> eigen(matrix(1:100,10,10))$values
>>>>  [1]  5.4e-311+ 0.0e+00i -2.5e-311+3.7e-311i -2.5e-311-3.7e-311i
>>>>  [4]  2.5e-312+ 0.0e+00i -2.4e-312+ 0.0e+00i  3.2e-317+ 0.0e+00i
>>>>  [7]   0.0e+00+ 0.0e+00i   0.0e+00+ 0.0e+00i   0.0e+00+ 0.0e+00i
>>>> [10]   0.0e+00+ 0.0e+00i
>>>>
>>>
>>> Mine is closer to Robin's, but not the same (EL4 x86).
>>>
>>>> eigen(matrix(1:100,10,10))$values
>>>   [1]  5.208398e+02+0.000000e+00i -1.583980e+01+0.000000e+00i
>>>   [3]  6.292457e-16+2.785369e-15i  6.292457e-16-2.785369e-15i
>>>   [5] -1.055022e-15+0.000000e+00i  3.629676e-16+0.000000e+00i
>>>   [7]  1.356222e-16+2.682405e-16i  1.356222e-16-2.682405e-16i
>>>   [9]  1.029077e-16+0.000000e+00i -1.269181e-17+0.000000e+00i
>>>>
>>>
>>> But surely, my matrix algebra is a bit rusty, I think this matrix is
>>> solveable analytically? Most of the eigenvalues shown are almost
>>> exactly zero, except the first two, actually, which is about 521
>>> and -16 to the closest integer.
>>>
>>> I think the difference between mine and Robin's are rounding errors
>>> (the matrix is simple enough I expect the solution to be simple
>>> integers
>>> or easily expressible analystical expressions, so 8 e-values being
>>> zero
>>> is fine). Peter's number seems to be all 10 e-values are zero or one
>>> being a huge number! So Peter's is odd... and Peter's machine also
>>> seems
>>> to be of a different archtecture (64-bit machine)?
>>>
>>> HTL
>>
>> Notice that Robin got something completely different in _R-devel_
>> which is where I did my check too.  In R 2.2.1 I get the expected two
>> non-zero eigenvalues.
>>
>> I'm not sure whether (and how) you can work out the eigenvalues
>> analytically, but since all columns are linear progressions, it is
>> at least obvious that the matrix must have column rank two.
>>
>
>
>
>
> For everyone's entertainment, here's an example where the analytic
> solution
> is known.
>
> fact 1:  the first eigenvalue of a magic square is equal to its constant
> fact 2: the sum of the other eigenvalues of a magic square is zero
> fact 3: the constant of a magic square of order 10 is 505.
>
> R-2.2.0:
>
> > library(magic)
> > round(Re(eigen(magic(10),F,T)$values))
> [1]  505  170 -170 -105  105   -3    3    0    0    0
> >
>
> answers as expected.
>
>
> R-devel:
>
>
>
> > a <- structure(c(68, 66, 92, 90, 16, 14, 37, 38, 41, 43, 65, 67, 89,
> 91, 13, 15, 40, 39, 44, 42, 96, 94, 20, 18, 24, 22, 45, 46, 69,
> 71, 93, 95, 17, 19, 21, 23, 48, 47, 72, 70, 4, 2, 28, 26, 49,
> 50, 76, 74, 97, 99, 1, 3, 25, 27, 52, 51, 73, 75, 100, 98, 32,
> 30, 56, 54, 80, 78, 81, 82, 5, 7, 29, 31, 53, 55, 77, 79, 84,
> 83, 8, 6, 60, 58, 64, 62, 88, 86, 9, 10, 33, 35, 57, 59, 61,
> 63, 85, 87, 12, 11, 36, 34), .Dim = c(10, 10))
>
> [no magic package!  it fails R CMD check !]
>
> > round(Re(eigen(magic(10),F,T)$values))
> [1] 7.544456e+165  0.000000e+00  0.000000e+00  0.000000e+00  0.000000e
> +00
> [6]  0.000000e+00  0.000000e+00  0.000000e+00  0.000000e+00  0.000000e
> +00
> >
>
>
> not as expected.
>
>
>
> --
> Robin Hankin
> Uncertainty Analyst
> National Oceanography Centre, Southampton
> European Way, Southampton SO14 3ZH, UK
>  tel  023-8059-7743
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>
>

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595



More information about the R-devel mailing list