Re: How Partial Simulations correctly determine non-halting ---Ben's 10/2022 analysis

On 6/7/24 11:16 PM, olcott wrote:

On 6/7/2024 9:43 PM, Mike Terry wrote:

On 07/06/2024 17:34, joes wrote:

Am Fri, 07 Jun 2024 10:40:37 -0500 schrieb olcott:

On 6/7/2024 10:32 AM, joes wrote:

Am Fri, 07 Jun 2024 10:20:09 -0500 schrieb olcott:

On 6/7/2024 10:09 AM, Python wrote:

Le 07/06/2024 à 17:05, olcott a écrit :

On 6/7/2024 9:55 AM, Python wrote:

Le 07/06/2024 à 16:47, olcott a écrit :

>

The issue here is that I proved that DD correctly simulated by HH
has different behavior than the directly executed DD(DD) and
everyone's "rebuttal" to this proof is to simply ignore it.

When you actually try to form a rebuttal of the above you will see
that I am correct. So far everyone simply ignores the proof that I
am correct as their only rebuttal.

A {correct simulation} means that each instruction of the above x86
machine language of DD is correctly simulated by HH and simulated in
the correct order.

And to the end. Thus it can't behave differently than direct execution.

You must not be very good at programming if you believe that
Infinite_Recursion must be simulated "to the end"

As I said before, if there is no end the simulation can't end either.

>
PO chooses to leave his terminology ambiguous, so we have 1000 post threads perpetuated by little more than verbal misunderstandings.
>
All PO's "simulating halt deciders" (SHD) are actually "partial simulating partial halt deciders":
They partially simulate their input computation, while monitoring the simulation state for patterns which *PO* just believes (without proof) indicate non-halting.  When such a pattern is detected, the SHD abandons its simulating activity ["aborts" its simulation], and returns nonhalting.  Also, PO does not claim that his SHDs decide all inputs correctly - or even that they halt for all inputs (which a decider must).  He is really just interested in the H(D,D) case.
>
PO has a couple of such rules/patterns, and at least one of them is sound: the "tight loop" pattern.  If an input (P,I) contains a tight loop (like "LOOP: goto LOOP"), and assuming PO has coded the rule correctly, his H/HH will detect the tight loop while simulating the code and correctly return non-halting.  In this scenario we have a non-terminating computation, but PO's /partial/ simulation obviously ends through aborting the target.  THERE'S NOTHING WRONG WITH THAT.  If someone wants to claim "that's not valid simulation" then right or wrong it's nothing more than a disagreement over use of terminology, and nothing is achieved by that.
>
PO also has at least one /unsound/ rule: his "infinite recursive simulation" rule.  That rule matches computations that in fact halt.

 The a sequence of configurations stops running because some
aspect of it has been abnormally terminated does not mean that
this sequence halts.
 Discussing this aspect is ridiculously too complex until after
you first understand that DD is correctly simulated by HH.
 We wasted three years on this because no one ever bothered
to see that I have conclusively proved that DD is correctly
simulated by HH at the x86 machine code level.
 

Despite PO's colourful language like "...until H 'sees that the simulation will never terminate unless blah blah'" and "...DD 'specifies infinite recursion to HH'" and so on, the truth is simply that his H is applying an unsound rule, which leads to it deciding incorrectly.)
>
And all the stuff about "correct simulations of DD by HH never reaching line nn" might be correct if suitably interpreted (as /partial simulations/ not reaching a simulated halt state, rather than the /actual computation/ being examined), but its all logically irrelevent having little to do with actual termination of a given computation.
>
What makes the SHD work or not work is the matching of rules (patterns detected in the simulated computation states) that the SHD uses to trigger an abort.  If a /sound/ rule matches, the SHD will decide correctly. If an /unsound/ rule matches it may decide incorrectly (as with PO's H(D,D)), or in other situations it might be correct "by fluke".
>
If PO's H used only sound rules, it would not decide input (D,D) incorrectly - but unfortunately neither would it decide correctly - it would never terminate /for that input/.  (I find that kind of neat...)
>

>

Anyone claiming that HH should report on the behavior of the directly
executed DD(DD) is requiring a violation of the above definition of
correct simulation.

But those are the same. How does simulating something change it?

It seems to boil down to you just don't know enough about programming
otherwise it would occur to you that there cannot possibly exist any
correct rebuttal to the following:

Answer the question. Why should the simulation differ from direct
execution?

>
[Let's assume code is properly restricted to exclude the complications of "cheating" in PO's x86utm environment through use of mutable static variables and the like.]
>
Obviously they wouldn't, except in so far as "partial simulation" differs from "full simulation". I.e. all steps will match precisely, up to the point where the partial simulation is abandoned.
>
And in PO's H(D,D) vs D(D) example that is exactly what the traces show:  The simulated and direct traces match exactly, right up to the point where H aborts the simulation.  (The full computation steps of D(D) then proceeding further until D terminates a bit later.)
>
Even though I put the two traces side-by-side for PO, he still insists the "behaviour is different".  Well, perhaps he just means that the H simulation was abandonned whereas directly executed D(D) proceeded further and eventually terminated?  If so, then /so what/?  I think his problem here is nothing to do with /simulations/ not reaching whatever line of code, but more the /reason/ that his H aborted - it matched his (unsound) non-halting pattern, and he just really really really really believes that pattern is sound.
>

 I have proven that DD correctly simulated by HH cannot possibly
stop unless aborted thus meeting this criteria:

No, you hve not, you have claimed and presented an argument. You have not PROVEN anything. It seems you don't understand what a Formal Proof is.

 Only these verbatim words were agreed to by MIT Professor Michael Sipser
<MIT Professor Sipser agreed to ONLY these verbatim words 10/13/2022>
   If simulating halt decider H correctly simulates its input D
   until H correctly determines that its simulated D would never
   stop running unless aborted then
    H can abort its simulation of D and correctly report that D
   specifies a non-halting sequence of configurations.
</MIT Professor Sipser agreed to ONLY these verbatim words10/13/2022>

Which is based on a DIFFERENT definition of "Correct Simulation" so its assable to you. He means a complete simulation that doesn't stop until it reaches a final state, which your H doesn't do (if it answer) nor does it correcly determine that an actual correct simulation of THIS input wouldn't halt.
Your arguement is based on thinking you are processing a template, as opposed to a program, so the input changes when you hypothetically change the HH processing THIS input, but the problem is NOT about deciding on a template, but on a program, which is an instance of that template based on a specific decider

 On 10/14/2022 7:44 PM, Ben Bacarisse wrote:
 > I don't think that is the shell game. PO really /has/ an H
 > (it's trivial to do for this one case) that correctly determines
 > that P(P) *would* never stop running *unless* aborted.
 

So now he needs to account for two "verified facts":  1)  D(D) computation never halts (/his unsound pattern matched!/)  2) D(D) computation halts when run directly.

 DD correctly simulated by HH never halts.
Directly executed DD(DD) is not the behavior
of the input that HH held accountable for.

But it is, if it is a Halt Decider. After all, the question a Halt Decider is answering is:
Does the Machine/Input described by the input halt when run?
Which *IS* about the directly execute DD(DD) so that *IS* what it is accountable for.
If you think it CAN'T do it, then you are just agreeing with the Halting Theorem that it can't be done.
Note, the question is not can we come up with something sort of like this that we can do, it is can we do this specific thing, that has a desired usage.

 I incorporate by reference
(a) The x86 language
(b) The notion of an x86 emulator
 (c) I provide this complete function

So?
What do you do with those?
Remember, a proof builds the chain from the Truth makers to your statement, explicitly showing the steps.
You never link from your "truths" to your conclusion.
And Proving that something doesn't happen is very hard, so claiming that NO HH can do something needs to actually be proven and not just claimed that since no one has found one it doesn't exist.

 void DDD(int (*x)())
{
   HH(x, x);
}
 _DDD()
[00001de2] 55         push ebp
[00001de3] 8bec       mov ebp,esp
[00001de5] 8b4508     mov eax,[ebp+08]
[00001de8] 50         push eax         ; push DD
[00001de9] 8b4d08     mov ecx,[ebp+08]
[00001dec] 51         push ecx         ; push DD
[00001ded] e890f5ffff call 00001382    ; call HH
[00001df2] 83c408     add esp,+08
[00001df5] 5d         pop ebp
[00001df6] c3         ret
Size in bytes:(0021) [00001df6]
 Then I state that No DDD correctly emulated by any
x86 emulator H can possibly reach its own [00001df6]
instruction.

Which needs to be PROVEN, not just stated.
Remember, H isn't "Just a simulator" but is DEFINED to be a decider that MUST return an answer.
So, your argument has it a contradiction in itself, telling you that you have a bad assumption somewhere.
Perhaps you can prove that it is impossible for any simulator to simulate to that point, but you also end up proving that behavior of DD (which is what is being asked about) MUST get there if HH meets its requirements. So, all you have done is proven that simulation can't be an answer to the problem.
Which has been a well known fact for decades, which is why you don't see much talk about them.

 

 That forces him to claim there are two different behaviours [/beyond/ the simple truncation of the aborted trace] for the one computation, and to come up with some Words to justify that position (like "pathological self reference", subjective questions) and then More Words to say that his other Words are "the true meaning of halting" etc. etc.  All nonsense flowing from not being able to accept his rule is just wrong.
>
>
Mike.
>