# Close reading: David H. Wolpert, "Implications of computer science theory for the simulation hypothesis" (arXiv:2404.16050v5)

Paper read: arXiv HTML v5, dated 19 Mar 2026, with checks against the arXiv source archive and bibliography. The same paper appears to have been published with the title "What computer science has to say about the simulation hypothesis" in *Journal of Physics: Complexity* 6(4), 045010 (2025), DOI 10.1088/2632-072X/ae1e50.

Retrieval note: I could retrieve the full arXiv HTML/source and IOP metadata. Google Scholar returned an automated-traffic block, PhilPapers returned a Cloudflare challenge, and Brave search was quota-limited. Semantic Scholar, Crossref, OpenAlex, IOP, arXiv, and INSPIRE checks are reported below.

## Q1. How exactly does Wolpert formalise the simulation hypothesis?

[Established from paper] Wolpert first restricts attention to what he calls computational universes. In section 2.2 he writes:

> "To connect with the PCT and CS theory more generally, I will consider universes that can be understood as evolving in discrete time, and that contain a subsystem that we will view as a 'computer'."

The state space is decomposed as `V = W x N`, where `W` is the environment and `N` is the computer:

> "where the set W, representing the environment, is initially finite and cannot be extended dynamically. On the other hand, N, representing the state of (IDs of) the computer, is countably infinite (or what's equivalent for current purposes, N is finite, but can be extended dynamically by arbitrary, finite amounts)."

The universe evolves deterministically by an evolution function:

> "I write the evolution of the universe from an initial, time-0 state (w0,n0) to a time t state as a vector-valued evolution function g of the initial state of that universe, g(t,w0,n0) = (wt,nt)."

He explicitly says he is not modeling whole cosmological universes in full generality:

> "Rather throughout this paper the expression 'universe' should be understand as shorthand for 'possibly coarse-grained subset of a universe'. However I do assume that the physical universes obey deterministic dynamics, be it over the original physical state space (assuming it is countable) or some coarse-grained version of such a state space." (section 2.2)

[Established from paper] His formal definition of simulation is Definition 1, section 2.3:

> "A universe V = W x N with evolution function g simulates the evolution of a universe V' = W' x N' with evolution function g' iff there exist three functions T(Delta t', w0', n0') in N, W(Delta t', w0', n0') in W, N(Delta t', w0', n0') in N such that for all Delta t' in N, w0' in W', n0' in N', t >= tau, g(t,omega,eta)_Nbar = <g'(Delta t', w0', n0')>, where as shorthand, tau := T(Delta t', w0', n0'), omega := W(Delta t', w0', n0'), eta := N(Delta t', w0', n0')."

The displayed equation is garbled by HTML conversion, but the content is clear from the source: after the simulating universe starts in `(omega, eta)`, the computer component of `V` eventually halts and contains an encoding of the simulated universe's state at the requested future time.

He adds immediately:

> "I say that V computably simulates V' if it simulates V', and in addition the three functions T(...), W(...), N(...) are all computable. Unless specified otherwise, whenever I refer to 'simulation' in this paper I implicitly assume it is computable."

[Established from paper] Simulation does not require the computer to run autonomously. Section 2.3:

> "Definition 1 does not require that the simulating computer N calculates the future state of the universe V' by itself, independently of the initial state of the environment W outside of N... The reason for this flexibility is to allow the computer N to retrieve the specific information it needs to perform its simulation... from some putative super-aliens who are running that computer N... Note though that this freedom also allows the beings running the simulation computer N to intervene on the dynamics of that computer at any time they want... They can even 'pull the plug early'."

[Established from paper] The host-computation/physics relationship is: the host universe must contain a computer subsystem `N`, and under RPCT it can implement arbitrary Turing-machine computations by appropriate initial conditions. The simulated universe, if it obeys PCT, has a computable evolution function. This is a syntactic formalism: section 1.2 says, "there is no distinction between universes that are 'real' and those that are 'only simulations'."

## Q2. How does he use the Physical Church-Turing thesis?

[Established from paper] Wolpert uses PCT as a formal definition for this paper, not as an argued empirical conclusion. Section 2.4:

> "Accordingly, for current purposes, I make the following (fully formal) definition: Definition 3. The Physical Church-Turing thesis (PCT) holds for universe V iff the evolution function g(.,.,.) of V is computable."

He then clarifies:

> "By Definition 3, if the PCT holds for V, there must be a UTM that (halts and) outputs the vector value of g(.,.,.), for all values of g's arguments."

And he warns that his Definition 3 may not match every historical use:

> "For these kinds of reasons, some readers might argue that Definition 3 doesn't exactly capture all of the various properties that have been referred to as the 'physical Church Turing thesis' in the literature... But for current purposes, it suffices to simply accept Definition 3, and ignore any objection such terminology might raise in certain researchers."

[Established from paper] He does not assume PCT for our actual universe. Section 1.2:

> "It is important to emphasize that in this paper I do not assume that the PCT applies to our actual physical universe. I do not even restrict attention to those universes that obey the laws of physics of our actual universe... I am only concerned with establishing the logical possibility of a physical system V that can (contain a computer that can) simulate some universe V'."

[Established from paper] He cites Piccinini/Maley, Piccinini, Copeland, Copeland/Shagrir, Aaronson, Arrighi & Dowek, Nielsen, Pour-El/Richards, Moore, Cubitt/Perez-Garcia/Wolf, Shiraishi/Matsumoto. The PCT subsection cites Arrighi & Dowek 2012 explicitly. I found no Gandy citation and no direct Deutsch citation in the bibliography/source. Deutsch-related quantum computation enters indirectly through quantum computing references, especially Nielsen/Chuang and Aaronson.

## Q3. The self-simulation result

[Established from paper] The theorem Wolpert uses is Kleene's second recursion theorem. Appendix B states it as:

> "Theorem 5. For any partial computable function Q(x,y) there is a Turing machine with index e such that T^e(x) = Q(x,e) for all x."

He then uses the total special case:

> "An important special case of the theorem is where Q(.,.) is a total computable function... I call this special case the 'total recursion theorem' in the main text."

[Established from paper] The self-simulation lemma itself is Lemma 2, section 4.1:

> "If both the PCT and the pristine RPCT hold for a universe V = W x N, then for all Delta t there exists n0 in N such that V freely simulates V for n0, Delta t."

The core fixed point is:

> "Since the PCT holds for V, its evolution function g(.,.,.) is computable (not just partial computable). Therefore if we fix Delta t... and then invoke the total recursion theorem, we see that there exists n* such that T^(n*)(w0) = g(Delta t, w0, n*) for all w0 in W, where both g(Delta t,w0,n*) and T^(n*)(w0) halt for all inputs w0."

After applying pristine RPCT, he concludes:

> "I will refer to Lemma 2 as the self-simulation lemma. It says that for any fixed Delta t, there is an associated initial state of the computer such that for any initial state of the rest of the universe, w0, that computer is guaranteed to halt and to output the state of the entire universe at time Delta t."

[Assessment for Argus] What Kleene buys him is a computable fixed point for the program loaded into the computer: an `n*` whose behavior on environment input `w0` returns the future state of the whole universe whose computer initially contains `n*`. This is stronger than merely saying "there is some fixed point"; he says `n0` is computable. Section 4.2:

> "Recall the convention that 'simulate' implicitly means 'computably simulate'... By definition, that means there is a halting program that constructs that n0."

[Assessment for Argus] The result is a claim of mathematical/logical possibility inside his formal framework, not a demonstrated physical possibility under our laws. Wolpert does distinguish this in section 1.2, quoted above: he is not investigating whether such a system exists "under the laws of physics as currently understood." His abstract phrase "we might be in a simulation being run on a computer by us" therefore means: if our universe satisfies the paper's assumptions (PCT plus pristine RPCT, deterministic coarse-grained dynamics, the formal decomposition `W x N`, etc.), then the formalism permits self-simulation. It is not an empirical claim that our laptops can do this.

## Q4. Rice's theorem impossibility results

[Established from paper] Section 8 restricts attention before applying Rice:

> "To illustrate these, for simplicity, throughout this subsection I'm restricting attention to universes with a countably infinite W as well as a countably infinite N."

The first Rice result fixes a universe `V` and asks which evolution functions/universes it can simulate:

> "First, Rice's theorem tells us that the set of all computational universes V' that can be simulated by a fixed universe V is undecidable... Define A(V) as the collection of all TMs that compute the evolution function of universes V' that are simulated by V... Therefore by Rice's theorem, it is undecidable whether an arbitrary (total TM and associated) V' is a member of A(V)."

The second fixes initial/final states and a simulation time:

> "As a variant of this result, again fix V, and also fix some spaces W', N'. Define B(V,(w0',n0'),(w_Delta t',n_Delta t')) as the collection of all TMs that compute the evolution function g' of some universe V'=W' x N' with the following two properties. First, g' sends (w0',n0') to (w_Delta t',n_Delta t'). Second, V simulates V' for the specific initial condition (w0',n0') and the simulation time Delta t. Again, it is undecidable whether an arbitrary TM lies in B(...)."

The third flips the direction:

> "Flipping things around, Rice's theorems tells us that for any fixed universe V', the set of all other universes V that can simulate V' is undecidable... An immediate consequence of this is that the set of all pairs of universes (V,V') such that V simulates V' is undecidable."

He then lists eight more undecidability claims:

1. "The set of universes (and in particular evolution functions) that obey the RPCT is undecidable."
2. "The set of universes that can simulate themselves is undecidable."
3. "The set of universes V that can simulate a universe V' != V that can in turn simulate any universe at all is undecidable."
4. "The set of universes V that can simulate a universe V' != V that can in turn simulate V is undecidable."
5. Restricting to self-simulating universes `V`, "the set of those universes in V that can simulate itself for all Delta t in N using some specific function S(Delta t) is undecidable."
6. For any `k >= 2`, the set of universes that can self-simulate for all `Delta t <= k` but not for some `Delta t > k` is undecidable.
7. The paper as rendered says: "for all Delta t >= k but not for some Delta t > k"; this appears internally inconsistent or a typo. UNCERTAIN: likely one inequality is intended differently.
8. "The set of those universes in V that can simulate itself for a time-ordered S(Delta t) is undecidable."

[Established from paper] Wolpert does make an epistemic-sounding conclusion after the Rice list:

> "There are some strange philosophical implications of these impossibility results, especially those that concern self-simulation. For example, it is possible that we are in a universe V that is simulating itself -- but only up to some future time, after which it is impossible for the simulation to still be accurate... The impossibility results above say that we can never be sure that this is not the case."

[Assessment for Argus] Does he prove "an embedded observer cannot determine X about whether it is simulated"? Partly, but not as a general observer-theory theorem. The formal theorem is undecidability of semantic properties of TMs/evolution functions. It says no algorithm decides these nontrivial properties over arbitrary indices. He also separately asserts, in section 6.1, that in exact self-simulation:

> "there would be no possible experimental test we could perform that could distinguish 'which of those two entities we are'."

and:

> "By the RPCT, there is no conceivable physical test, no observable value, that could tell us which of those two dynamic processes 'is' us."

Those statements are about indistinguishability between duplicated exact instances in a self-simulation scenario, not a Rice theorem about deciding "am I simulated?" from finite observations.

[Assessment for Argus/H17] Wolpert does not make H17 a theorem. He does not prove that the generic simulation hypothesis has no assignable likelihood because designer aims/abilities are unspecified. He does argue adjacent probability problems: section 1.1 says probability assignments over universes "beg (many) questions" and cannot be experimentally assessed with proper scoring rules; section 2.4 says no uniform probability distribution exists over the countably infinite class of finite-`W` PCT universes; section 7 says a Cantor measure cannot be assigned to a natural simulation graph if the simulation relation is used to fix the measure. These support parts of Argus's anti-generic-probability posture, but by a different route from Sober and not at H17's full generality.

[Assessment for Argus/H18] Wolpert does not settle H18. He assumes computability for PCT universes, mentions theoretical uncomputable physical systems and PCT disputes, and uses Rice to show undecidability of properties of Turing-computable evolution functions. I found no theorem about whether a finite embedded observer can certify from finite finite-precision measurements that a physical process computes a non-Turing-computable function.

[Assessment for Argus] I found no observation in Wolpert that contradicts H17 by assigning a likelihood to the generic simulation hypothesis or giving an observation that bears on it generically. His framework gives sufficient conditions for simulation/self-simulation, not a Bayesian evidential model.

## Q5. Fully homomorphic encryption section

[Established from paper] Section 6.2 considers a simulator running an FHE version of the simulated universe's evolution function:

> "Another interesting set of issues arises if there is one universe V = W x N that simulates a second universe V' = W' x N', but that simulation is a fully homomorphic encrypted (FHE) version of the evolution function of that second universe."

His explanatory footnote says:

> "Recall that in FHE you have an algorithm that runs on some encrypted data, producing a result that when decrypted is identical to the result you would get if ran an associated algorithm on the original, pre-encrypted data, without any encryption of any sort... the program n0 in N could be an FHE version of the program simulating V' that the beings who are running that computer N want it to compute. So those beings would need to use a decryption key to understand the result of their computer's simulation of the evolution of V'."

The consequences are:

> "if the simulator beings lost the decryption key, then they would not be able to read out the results of their simulation. This would be the case even though that simulation would in fact be perfectly accurate."

For inhabitants:

> "since V' obeys the PCT there is no sense in which the beings being simulated could know that they are being produced in a simulation. Life would appear 'normal' to them, with no 'randomness' of any sort. So in particular, there is no way that they would be able to distinguish between being produced in a simulation being made via an FHE algorithm, or instead in some simulation that is easier to understand."

He adds a striking external-perspective claim:

> "if we could actually somehow see our universe's evolution from the perspective of the beings who are running the program producing us, we would not be able to distinguish the laws of physics controlling our universe in that simulation from completely random noise. In this sense, the actual laws of physics in our universe might in fact be pure noise -- and we would not be able to tell the difference."

[Assessment for Argus] The concrete claim is: FHE can hide the simulator's computation from the simulator beings without altering the simulated inhabitants' physics. It does not give inhabitants an empirical handle on whether encryption is used. It also decouples apparent randomness of the host-level representation from low Kolmogorov complexity of the underlying simulated law.

## Q6. What does Wolpert say is not implied / left open?

[Established from paper] He explicitly declines to assume or prove PCT/RPCT for our universe. Section 1.2:

> "I do not assume that the PCT applies to our actual physical universe. I do not even restrict attention to those universes that obey the laws of physics of our actual universe... I do not investigate the possibility of such a system under the laws of physics as currently understood -- never mind the even more narrowly defined question of whether there is such a physical system in our universe."

[Established from paper] Definition 3 does not cover super-Turing host universes:

> "the analysis below does not apply to any universes so many levels of computational power above our own that they can contain computers capable of super-Turing computation... that are simulating our universe." (section 2.4)

[Established from paper] RPCT is not implied by PCT:

> "even if the PCT holds in our specific universe, that does not mean that the RPCT has to hold as well. In particular, cosmological considerations could prevent it from holding." (section 2.5)

[Established from paper] He leaves computational complexity mostly open. Section 9:

> "There are many mathematical questions suggested by the self-simulation lemma that I am not considering in this paper. Most obviously, I have not considered the computational complexity of finding S(Delta t), and its dependence on g, Delta t, |W|, etc."

He also leaves open minimal simulation time, density of nested self-simulation times, approximate simulation, average-case complexity, time-minimal and space-minimal simulation graphs.

[Established from paper] He leaves quantum/relativistic extensions open. Section 10:

> "it might be worth investigating extensions of the analysis in this paper to concern quantum mechanical and / or relativistic universes. One obvious question in this regard is whether the quantum no-cloning theorem means that self-simulation could never arise (at the quantum level) in our universe."

[Established from paper] Appendix C admits a possible subtlety about hard-coding `Delta t` and multiple fixed points:

> "It is not clear though that there is a way to exploit this flexibility so that there is at least one n* that satisfies the recursion theorem for all Delta t. So for simplicity, this possibility is not considered in this paper."

## Q7. Who does he cite, and who has cited him?

[Established from paper/source] Important cited references include:

- Bostrom 2003, "Are we living in a computer simulation?"
- Chalmers 2022, *Reality+*
- Beane, Davoudi & Savage 2014, "Constraints on the Universe as a numerical simulation"
- Campbell, Owhadi, Sauvageau & Watkinson 2017, "On testing the simulation theory"
- Barrow 2007/2011
- Hanson 2001; Weatherson 2003; Dainton 2012; Conitzer 2019; Lim 2022
- Aaronson 2005, 2013; an Aaronson 2024 blog quotation in the text
- Arrighi & Dowek 2012, "The physical Church-Turing thesis and the principles of quantum theory"
- Piccinini 2011; Piccinini & Maley 2010/2021; Copeland 2023; Copeland & Shagrir 2018
- Lloyd 1992, 2000, 2013
- Tegmark 1998, 2008; Hut, Alford & Tegmark 2006 in the journal references
- Kleene 1952; Moschovakis 2010; Sipser; Arora & Barak; Li & Vitanyi
- Pour-El & Richards 1982; Moore 1990/1991; Cubitt et al. 2015; Shiraishi & Matsumoto 2021
- Wolfram 1984; Codd 1968; Israeli & Goldenfeld 2004; Mitchell 1996
- Zurek 1989; Caves 1990/1993; Baez & Stay 2012

[Established from source search] I found no Gandy citation. I found no Zuse citation. I found no live Schmidhuber citation; "schmidhuber" appears only in a commented-out source line about Kolmogorov complexity/Occam material, not in the bibliography. I found no direct Deutsch citation.

[Established from metadata] Publication status: yes, journal publication found. Crossref/IOP list: David H. Wolpert, "What computer science has to say about the simulation hypothesis," *Journal of Physics: Complexity* 6(4), 045010, online 2025-12-01, DOI 10.1088/2632-072X/ae1e50. The arXiv title differs from the journal title, but the abstract and source filename (`post.publication`) indicate this is the journal version lineage.

[Established from metadata, with caveats] Citation checks:

- Semantic Scholar for arXiv:2404.16050 reports 3 citations: (1) Salman Sajad Wani et al., "Constructive realization of self-referential prediction limits in quantum control: resource bounds and Godel-safe architectures" (2026); (2) David H. Wolpert & Jordan Scharnhorst, "Stochastic Process Turing Machines" (2024); (3) David H. Wolpert, "What can we know about that which we cannot even imagine?" (2022). The latter two are not straightforward post-publication critiques/replies.
- Semantic Scholar for DOI 10.1088/2632-072X/ae1e50 reports 1 citation: Eran Kopel, "Certified coherent, informative, and non-entanglement-breaking fixed points of future-referential quantum feedback" (2026).
- OpenAlex reports the journal article with cited_by_count = 1, but its cited-by endpoint returned no concrete works in my query; treat this as index inconsistency.
- Crossref reports is-referenced-by-count = 0 for the DOI.
- INSPIRE query for the exact title returned no visible record.
- Google Scholar could not be checked because it returned an automated-traffic page. PhilPapers could not be checked because it returned a Cloudflare challenge.

[Assessment] I found no published critique or direct reply in the accessible checks. The known citations look like downstream technical papers, not critiques of Wolpert's argument.

## Commentary for Argus

Wolpert's paper is most useful to Argus as a formal-CS sharpening of what "simulation" can mean under PCT/RPCT, not as an evidential/Bayesian treatment of the generic simulation hypothesis. It strengthens Argus's skepticism about generic probabilities in a local way: no uniform distribution over the relevant countable class, no natural Cantor measure under the simulation-graph relation, and self-simulation makes counting "instances of us" ill-posed. But it does not derive H17 in Sober's likelihood sense.

The self-simulation claim should be read as: a computable fixed point exists in a deterministic, coarse-grained, PCT/RPCT universe formalism. It is not a demonstration that our actual universe satisfies those premises.