如果您将列表复制到数组中,则以下内容可能很有用:由于我们仅考虑偶数长度回文,因此我假设这种情况。但是该技术可以轻松扩展以处理奇数长度回文。
我们存储的不是回文的实际长度,而是长度的一半,因此我们知道可以向左/右移动多少个字符。
考虑单词:
aabbabbabab
。我们正在寻找最长的回文。
a a b b a b b a b a b (spaces for readability)
°^° start at this position and look to the left/right as long as possible,
1 we find a palindrome of length 2 (but we store "1")
we now have a mismatch so we move the pointer one step further
a a b b a b b a b a b
^ we see that there's no palindrome at this position,
1 0 so we store "0", and move the pointer
a a b b a b b a b a b
° °^° ° we have a palindrome of length 4,
1 0 2 so we store "2"
naively, we would move the pointer one step to the right,
but we know that the two letters before pointer were *no*
palindrome. This means, the two letters after pointer are
*no* palindrome as well. Thus, we can skip this position
a a b b a b b a b a b
^ we skipped a position, since we know that there is no palindrome
1 0 2 0 0 we find no palindrome at this position, so we set "0" and move on
a a b b a b b a b a b
° ° °^° ° ° finding a palindrome of length 6,
1 0 2 0 0 3 0 0 we store "3" and "mirror" the palindrome-length-table
a a b b a b b a b a b
^ due to the fact that the previous two positions hold "0",
1 0 2 0 0 3 0 0 0 we can skip 2 pointer-positions and update the table
a a b b a b b a b a b
^ now, we are done
1 0 2 0 0 3 0 0 0 0
这意味着:一旦我们找到回文位置,就可以推断出表的某些部分。
另一个例子:
aaaaaab
。
a a a a a a b
°^°
1
a a a a a a b
° °^° °
1 2 1 we can fill in the new "1" since we found a palindrome, thus mirroring the
palindrome-length-table
a a A A a a b (capitals are just for emphasis)
^ at this point, we already know that there *must* be a palindrome of length
1 2 1 at least 1, so we don't compare the two marked A's!, but start at the two
lower-case a's
我的观点是:一旦我们找到回文,我们就可以镜像(至少部分)回文长度表,并推断出有关新字符的信息。这样,我们就可以节省比较的时间。