diff options
author | enge <enge@211d60ee-9f03-0410-a15a-8952a2c7a4e4> | 2010-06-16 12:36:26 +0000 |
---|---|---|
committer | enge <enge@211d60ee-9f03-0410-a15a-8952a2c7a4e4> | 2010-06-16 12:36:26 +0000 |
commit | 0d50d1bca19dd6efd5285cdf2026b07b9e586ad5 (patch) | |
tree | 33d897c100260e08ab5e49c60323e35b6cf6f6e8 /doc | |
parent | c40c996f085fa0a98fc680268167abf0a00d5792 (diff) | |
download | mpc-0d50d1bca19dd6efd5285cdf2026b07b9e586ad5.tar.gz |
algorithms.tex: removed assumption on "same quadrant" when applying Prop. 9,
which has been dropped in r744
git-svn-id: svn://scm.gforge.inria.fr/svn/mpc/trunk@782 211d60ee-9f03-0410-a15a-8952a2c7a4e4
Diffstat (limited to 'doc')
-rw-r--r-- | doc/algorithms.tex | 8 |
1 files changed, 3 insertions, 5 deletions
diff --git a/doc/algorithms.tex b/doc/algorithms.tex index ab25682..850d9a8 100644 --- a/doc/algorithms.tex +++ b/doc/algorithms.tex @@ -46,7 +46,7 @@ \title {MPC: Algorithms and Error Analysis} \author {Andreas Enge \and Philippe Th\'eveny \and Paul Zimmermann} -\date {Draft; March 30, 2010} +\date {Draft; June 16, 2010} \begin {document} \maketitle @@ -617,8 +617,7 @@ then $\epsilon_1 \epsilon_2 \leq 2^{1-p}$ and $\epsilon \leq \big( \max (k_{R, 1}, k_{I, 1}) + \max (k_{R, 2}, k_{I, 2}) + 1 \big) 2^{1 - p}$. Applying Propositions~\ref {prop:comrelerror} and~\ref {prop:relerror} -in the converse direction yields, under the assumption that $\corr z$ -and $\appro z$ lie in the same quadrant of the complex plane, +in the converse direction yields \begin {equation} \label {eq:propmulcomrel} \begin {array}{rl} @@ -838,8 +837,7 @@ with \epsilon \leq (|\theta_1| + |\theta_2|) \sum_{k = 0}^\infty |\theta_2|^k. \] Using the same notation and assumptions as at the end of -\S\ref {sssec:propmul}, in particular that $\corr z$ and $\appro z$ -lie in the same quadrant and that all higher order error terms +\S\ref {sssec:propmul}, in particular that all higher order error terms (involving $\epsilon_1^2$, $\epsilon_2^2$, $\epsilon_1 \epsilon_2$ or higher powers) are absorbed by $2^{\Exp (\appro x) - p}$, we find the exact same error estimate \eqref {eq:propmulcomrel} |