A survey on different triangular norm-based fuzzy logics (Q1299619): Difference between revisions

The authors analyze properties of \([0,1]\)-based logic which use Frank's t-norm \(x \& y=\log_\lambda(1+(\lambda^x-1)(\lambda^y-1)/(\lambda-1))\), \(0\leq\lambda\leq\infty\) as an ``and'' operation. There are two ways to expand \(\&\) to full logic: in the R-implication approach, we define \(x\to y=\sup\{z |x\&z\leq y\}\), \(\neg x\) as \(x\to 0\), and \(x\vee y\) as \(\neg(\neg x \& \neg y)\). Since all strictly Archimedean t-norms are isomorphic, the logics corresponding to \(0<\lambda<\infty\) are equivalent to the logic corresponding to \(\lambda=1\) and \(x \& y=x\cdot y\). Thus, we only need to consider three logics corresponding to \(x\cdot y\), \(\min(x,y)\) (\(\lambda=0\)), and \(\max(x+y-1,0)\) (\(\lambda=\infty\)). For all three cases, the authors prove completeness results and a weak deduction theorem -- that \(T\cup \{\varphi\}\vdash \psi\) iff \(\exists n (T\vdash \varphi \& \ldots \& \varphi\;(n\text{\;times})\to\psi)\). For \(\lambda=\infty\), a strong deduction theorem is also true: \(T\cup \{\varphi\}\vdash \psi\) implies \(T\vdash \varphi\to\psi\). The resulting R-logic has nice logical properties, but it also has serious counter-intuitive features: e.g., \(\neg x\) is always crisp. A more intuitive version is S-logic, where \(\neg x=1-x\), \(x\vee y=\neg(x \& \neg y)\), and \(x\to y\) is defined as \(y\vee \neg x\). These S-logics have some nice logical properties, but not as many as R-logics. It is worth mentioning that although the authors concentrate on Frank's logics, (due to the fact that all strictly Archimedean t-norms are equivalent) their results on R-logics with \(0<\lambda<\infty\) hold for all R-logics stemming from a strictly Archimedean t-norm.