This function is not yet fully documented. This is a transcript of the text-formatted help.
bbtauest Estimate accuracy of a black-box operator. TAU=bbtauest(M) approximates the accuracy of the matrix-vector products M*x and M'*x, when x is a unit vector. TAU=bbtauest(M,K) uses 2*K function evaluations. [TAU,TAUAMP]=bbtauest(M) also approximates the amplification of previous errors, i.e. the 2-norm of M. bbtauest will also determine suitable parameters for a normal Matlab matrix (full or sparse). This is handy when matrices are used as building-blocks. The routine makes no assumptions about the values of tau and tauamp of the operator. Therefore, it can be used to estimate the values for an operator when a rounding error analysis is unavailable. bbtauest offers a way to test that an operator implements the adjoint correctly. If is usually able to point out if there is an algorithmic mistake in the implementation of the product A'*X. The ratio tau/tauamp should not be too large. In most cases it should be less than 1e-10, and it should always be less than 1. Note: estimating the rounding errors is a difficult process. An analytical result should be used if possible. See also , bbcompare, .