RW wrote:Roy McCoy wrote:I have a question about the BUG+1 example at Sudopedia, where it says:
As noted above, the same deduction can be made by coloring. Specifically, the chain
r1c3=9 => r7c3=1 => r7c5=3 => r6c5=9
leaves no candidates for digit 9 in box 2.
As noted above where? And while I can see how coloring with the indicated chain can also provide a solution for the puzzle, how is it "the same deduction"? If the 9 can't be at r1c3, is it immediately obvious why it couldn't be at either r1c4 or r1c5? If so, how?
You're right, it's not a very good example. It's not the same deduction neither is it immediately clear that r1c4 or r2c5 cannot be 1. But if we really really want to use that chain to make the same deduction, then we can note that the chain works both ways:
r1c3=9 => r7c3=1 => r7c5=3 => r6c5=9
r6c5=9 => r7c5=3 => r7c3=1 => r1c3=9
In other words, neither r1c3 or r6c5 can be 9. From this we can deduce that r1c5 must be 9.
Though I don't think this is what the author of the Sudopedia article had in mind.
RW
If we express that chain in full we find every inference is conjugate (one of the two arguments must be true and the other false).
(9'~1")r1c3 ~ (1')r7c3 ~ (1"~3')r7c5 ~ (3"~9')r6c5
Either all the candidates tagged 0' are true or all those tagged 0" are, and so (9)r1c3 and (9)r6c5 are equivalent. If they are assumed to be true a contradiction is produced so they must be false.
So
Roy what you describe as a 'ambidirectional chain' in Sudopedia is a conjugate chain expressed in forcing chain notation. As such it's a bit of a dog's dinner in my eyes.
Purists would consider this deduction to be assumptive because it shows a contradiction if the end nodes are assumed to be true. To satisfy them there are two AICs to make the same eliminations individually:
(5=9)r1c4 - (9=1)r1c3 - (1)r7c3 = (1)r7c5 - (1=5)r8c5 => r1c3 <> 5, r8c4 <> 5
(7=1)r1c9 - (1)r1c3 = (1)r7c3 - (1=3)r7c5 - (3=9)r6c5 - (9=7)r2c5 => r1c5 <> 7
These eliminations aren't considered assumptive because the end arguments can't both be false and no truth state has to be assumed.
The first of these AICs also solves the BUG+1 pattern as it reduces r8c4 to a singleton.
RW I agree that what the author of the article had in mind when he said that colouring could be used is obscure, and probably not that chain.