home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math
- Path: sparky!uunet!snorkelwacker.mit.edu!galois!nevanlinna!jbaez
- From: jbaez@nevanlinna.mit.edu (John C. Baez)
- Subject: Re: Help - non-integral power of a matrix?
- Message-ID: <1992Aug12.231708.3644@galois.mit.edu>
- Keywords: matrix
- Sender: news@galois.mit.edu
- Nntp-Posting-Host: nevanlinna
- Organization: MIT Department of Mathematics, Cambridge, MA
- References: <Aug.10.15.45.34.1992.26563@clam.rutgers.edu> <carle.713599691@vex> <a_rubin.713653963@dn66>
- Date: Wed, 12 Aug 92 23:17:08 GMT
- Lines: 40
-
- In article <a_rubin.713653963@dn66> a_rubin@dsg4.dse.beckman.com (Arthur Rubin) writes:
- >In <carle.713599691@vex> carle@vex.ugcs.caltech.edu (Matthew Thomas Carle) writes:
- >
- >>gonzalez@clam.rutgers.edu (Ralph Gonzalez) writes:
- >
- >
- >>>Hi. Does anyone know of an algorithm to find a non-integral
- >>>power of a matrix, e.g. A^.5 or A^1.3? Thus, A^2 is the same
- >>>as AxA and A^0 is the identity.
- >
- >>>I imagine if such a thing is defined, then there are conditions
- >>>on A...
-
- >log "obviously" converges if all eigenvalues of A are strictly within 1 of 1,
-
- To clarify, perhaps, let me add that this is not only "obvious", it's
- true, at least if A is diagonalizable. If A is a
- not-necessarily-diagonalizable matrix log A is defined if ||A - 1|| < 1;
- I don't think the eigenvalue condition above is sufficient.
-
- Gonzalez was interested one-parameter groups of matrices. Well, that's
- not how he phrased it, but he was looking for matrix-valued functions of
- t such that A(t)A(s) = A(t+s). A^t is a nice way to think about these
- heuristically but it's sort of a nuisance because fractional powers of
- matrices are even more multivalued and branchy than fractional powers of
- numbers. If A is close enough to 1 (as made precise above), there is
- a unique "best" definition of log A, namely
-
- (A-1) - (A-1)^2/2 + (A-1)^3/3 - ...
-
- the usual power series. One can see that this satisfies exp(log A) = A
- and that exp(t log A) is good way of defining A^t, that is, it's a
- one-parameter group.
-
- In general, though, it is best to specify one-parameter groups by their
- generators. That is, any (continuous) 1-parameter group of matrices is
- of the form exp(t C) for some matrix C. If the 1-parameter group is a
- rotations then C is an "infinitesimal rotation" (a skew-adjoint matrix)
- and so on... huge amounts of stuff are known about this, so that's how I
- would advise Gonzalez to proceed.
-