אִם יִרְצֶה הַשֵּׁם
The string is not what vibrates. The incompleteness of the string is what vibrates. That vibration is the only thing left that can leave an empirical scratch on reality.
The vibrational modes are mathematical structure. Their imprint on observables passes through a filter (compactification, moduli stabilisation, thermalisation) that is provably uncomputable in general. What reaches us is the residue of that filter, the narrow band of cases where the filter halts with a yes. The shape of that residue, clustered rather than smooth, sharp rather than tailed, is observable.
That is not philosophy. It leaves fingerprints.
Peter Woit kicked the dog again with a post titled "String Theory is Not the Theory of the Real World." He riffed on Susskind's recent admissions about the program's predictive emptiness (Not Even Wrong). The usual sociology followed. I want to take the criticism seriously, but filter it through the meta-theoretical framework Faizal, Krauss, Shabir, and Marino laid out in JHAP 2025, "Consequences of Undecidability in Physics on the Theory of Everything." Put the two together, and the $10^{500}$ vacua problem stops looking like a statistical embarrassment and starts looking like a computability filter.
Scope
This is about cosmology. String theory has earned its keep elsewhere. Hadronic resonances below 1 GeV and black hole microstate counting (Strominger-Vafa 1996) are clear wins. Both are domains where the relevant question admits a halting answer and the math delivers. The conjecture below is not about those regimes. It is about the cosmological application, the part where you need the 10-D or 26-D spectrum to look like a 4-D hot universe with a CMB. That is where Woit's complaint bites and where I want to push.
The undecidability lever
Faizal et al push a claim that should make anyone working in foundations sit up. Several physical questions of direct interest are formally undecidable. Not NP-hard. Undecidable in the Turing sense. The headline case is the spectral-gap problem (Cubitt, Perez-Garcia, Wolf, Nature 2015). No algorithm takes a 2D Hamiltonian description and returns yes or no on whether the system is gapped. Shiraishi and Matsumoto (Nature Communications 2021) extended this directly into thermalisation. Whether a quantum system thermalises is similarly undecidable. Faizal et al cite exactly this result inside their no-go on algorithmic theories of everything.
Gap controls thermalisation. Whether a many-body system equilibrates, whether it admits a hot dense early epoch that cools to a cold sparse late epoch, all of it rides on gap structure and on the dynamical equilibration question. Both are undecidable in general.
The conjecture
To make a 10-D superstring spectrum look like 4-D hot matter, you have to thermalise the moduli. The internal degrees of freedom must dump energy into the visible sector and produce a hot Friedmann phase. This is not optional. Without it you get no recombination surface, no acoustic peaks, and no observers.
But whether a given compactification thermalises is exactly the kind of question Cubitt et al and Shiraishi-Matsumoto showed is undecidable. The connection to strings is direct. Faizal, Shabir, and Khan (Nucl. Phys. B 1010, 2025) pushed Goedel-style limits into string field theory and group field theory. The vacuum set is not a flat list of $10^{500}$ items you can statistically sample. The map from compactification data to thermalisation outcome is uncomputable as a function.
How does this act as a physical filter? Think about eternal inflation. Vacua are populated by bubble nucleation events. Those amplitudes depend on the moduli potential's local features near the nucleation site. If recovering those features from the compactification specification reduces to a halting decision on the underlying Hamiltonian, then nucleation amplitudes for non-halting specifications lack a defining limit. They do not just fail to thermalise. They fail to nucleate. The anthropic patch is the recursively enumerable subset that admits a halting certificate.
Heading off the obvious objection
Blumberg made the right objection in January 2026. A proof-theoretic limit is not a dynamical limit. Goedel and Tarski constrain what a formal system can prove inside its own language. They do not stop a process from running. The cosmos does not need to halt a Turing machine to evolve.
Granted. Uncomputable vacua evolve fine. Local quantum dynamics runs whether or not anyone can prove what it is doing.
But anthropic selection is an inferential claim. It requires assigning probabilities to outcomes conditioned on compactification data. A function can be perfectly well-defined as a measure-theoretic object while being uncomputable as a procedure. The anthropic prior might exist as a formal measure, but the conditioning relation cannot be approached by any uniform algorithm. The inferential content breaks down. You cannot say our universe is typical among observer-supporting vacua because the denominator is uncomputable, even with infinite computational time. It is a projection onto whatever subset of specifications admits a halting answer.
What this would predict
Statistical selection over $10^{500}$ random vacua predicts smooth posteriors over dimensionless parameters. Sharper measurements narrow the posteriors but keep them smooth.
A computability filter places observables on points of low Kolmogorov complexity in their natural specification language. Halting configurations cluster near specifications with short halting certificates. Observables should not be drawn from smooth distributions. They should cluster near discrete attractor values.
Three concrete signatures.
CMB spectral index running. Current Planck and ACT data are consistent with zero running of $dn_s/d\ln k$ at the $10^{-3}$ level. A smooth prior says the next decade tightens the posterior toward a centroid. A computability filter says the value should sit at a preferred discrete point tied to a low-complexity inflationary potential. CMB-S4 and LiteBIRD can discriminate.
Coupling-constant correlations. A smooth prior says deviations of $\alpha_{\text{em}}$, $\alpha_s$, and lepton mass ratios from naive predictions are uncorrelated. A computability filter says they should be correlated. A single short halting certificate constrains many constants simultaneously. This is the strongest signature because correlation tests are statistically powerful.
CMB spectral distortions. If thermalisation is a halt condition rather than a smooth limit, cosmologies near the boundary of the computable subset should show incomplete thermalisation. The $\mu$ and $y$ distortion parameters measure exactly this. A smooth prior predicts no preferred scale beyond ordinary dissipation physics. A computability filter predicts a characteristic scale tied to halting structure. PIXIE-class missions can hit the range.
There is also a negative signature. Statistical priors predict observable tails like rare coincidences at calculable rates. Computability filters are sharp at the boundary of the decidable set. Persistent absence of statistical anomalies that anthropic priors predict is itself evidence for a sharp filter.
References
Cubitt, T., Perez-Garcia, D., Wolf, M. (2015). "Undecidability of the spectral gap." Nature 528, 207-211.
Faizal, M., Krauss, L. M., Shabir, A., Marino, F. (2025). "Consequences of Undecidability in Physics on the Theory of Everything." Journal of Holography Applications in Physics 5(2), 10-21. DOI: 10.22128/jhap.2025.1024.1118.
Faizal, M., Shabir, A., Khan, A. K. (2025). "Consequences of Goedel theorems on third quantized theories like string field theory and group field theory." Nuclear Physics B 1010, 116774.
Shiraishi, N., Matsumoto, K. (2021). "Undecidability in quantum thermalization." Nature Communications 12, 5084.
Strominger, A., Vafa, C. (1996). "Microscopic origin of the Bekenstein-Hawking entropy." Phys. Lett. B 379, 99-104.
Blumberg, M. (2026). "Undecidability Does Not Kill Simulation." https://www.svgn.io/p/undecidability-does-not-kill-simulation
Woit, P. (2025). "String Theory is Not the Theory of the Real World." Not Even Wrong. https://www.math.columbia.edu/~woit/wordpress/?p=14059
אִם יִרְצֶה הַשֵּׁם
The string is not what vibrates. The incompleteness of the string is what vibrates. That vibration is the only thing left that can leave an empirical scratch on reality.
The vibrational modes are mathematical structure. Their imprint on observables passes through a filter (compactification, moduli stabilisation, thermalisation) that is provably uncomputable in general. What reaches us is the residue of that filter, the narrow band of cases where the filter halts with a yes. The shape of that residue, clustered rather than smooth, sharp rather than tailed, is observable.
That is not philosophy. It leaves fingerprints.
Peter Woit kicked the dog again with a post titled "String Theory is Not the Theory of the Real World." He riffed on Susskind's recent admissions about the program's predictive emptiness (Not Even Wrong). The usual sociology followed. I want to take the criticism seriously, but filter it through the meta-theoretical framework Faizal, Krauss, Shabir, and Marino laid out in JHAP 2025, "Consequences of Undecidability in Physics on the Theory of Everything." Put the two together, and the $10^{500}$ vacua problem stops looking like a statistical embarrassment and starts looking like a computability filter.
Scope
This is about cosmology. String theory has earned its keep elsewhere. Hadronic resonances below 1 GeV and black hole microstate counting (Strominger-Vafa 1996) are clear wins. Both are domains where the relevant question admits a halting answer and the math delivers. The conjecture below is not about those regimes. It is about the cosmological application, the part where you need the 10-D or 26-D spectrum to look like a 4-D hot universe with a CMB. That is where Woit's complaint bites and where I want to push.
The undecidability lever
Faizal et al push a claim that should make anyone working in foundations sit up. Several physical questions of direct interest are formally undecidable. Not NP-hard. Undecidable in the Turing sense. The headline case is the spectral-gap problem (Cubitt, Perez-Garcia, Wolf, Nature 2015). No algorithm takes a 2D Hamiltonian description and returns yes or no on whether the system is gapped. Shiraishi and Matsumoto (Nature Communications 2021) extended this directly into thermalisation. Whether a quantum system thermalises is similarly undecidable. Faizal et al cite exactly this result inside their no-go on algorithmic theories of everything.
Gap controls thermalisation. Whether a many-body system equilibrates, whether it admits a hot dense early epoch that cools to a cold sparse late epoch, all of it rides on gap structure and on the dynamical equilibration question. Both are undecidable in general.
The conjecture
To make a 10-D superstring spectrum look like 4-D hot matter, you have to thermalise the moduli. The internal degrees of freedom must dump energy into the visible sector and produce a hot Friedmann phase. This is not optional. Without it you get no recombination surface, no acoustic peaks, and no observers.
But whether a given compactification thermalises is exactly the kind of question Cubitt et al and Shiraishi-Matsumoto showed is undecidable. The connection to strings is direct. Faizal, Shabir, and Khan (Nucl. Phys. B 1010, 2025) pushed Goedel-style limits into string field theory and group field theory. The vacuum set is not a flat list of $10^{500}$ items you can statistically sample. The map from compactification data to thermalisation outcome is uncomputable as a function.
How does this act as a physical filter? Think about eternal inflation. Vacua are populated by bubble nucleation events. Those amplitudes depend on the moduli potential's local features near the nucleation site. If recovering those features from the compactification specification reduces to a halting decision on the underlying Hamiltonian, then nucleation amplitudes for non-halting specifications lack a defining limit. They do not just fail to thermalise. They fail to nucleate. The anthropic patch is the recursively enumerable subset that admits a halting certificate.
Heading off the obvious objection
Blumberg made the right objection in January 2026. A proof-theoretic limit is not a dynamical limit. Goedel and Tarski constrain what a formal system can prove inside its own language. They do not stop a process from running. The cosmos does not need to halt a Turing machine to evolve.
Granted. Uncomputable vacua evolve fine. Local quantum dynamics runs whether or not anyone can prove what it is doing.
But anthropic selection is an inferential claim. It requires assigning probabilities to outcomes conditioned on compactification data. A function can be perfectly well-defined as a measure-theoretic object while being uncomputable as a procedure. The anthropic prior might exist as a formal measure, but the conditioning relation cannot be approached by any uniform algorithm. The inferential content breaks down. You cannot say our universe is typical among observer-supporting vacua because the denominator is uncomputable, even with infinite computational time. It is a projection onto whatever subset of specifications admits a halting answer.
What this would predict
Statistical selection over $10^{500}$ random vacua predicts smooth posteriors over dimensionless parameters. Sharper measurements narrow the posteriors but keep them smooth.
A computability filter places observables on points of low Kolmogorov complexity in their natural specification language. Halting configurations cluster near specifications with short halting certificates. Observables should not be drawn from smooth distributions. They should cluster near discrete attractor values.
Three concrete signatures.
CMB spectral index running. Current Planck and ACT data are consistent with zero running of $dn_s/d\ln k$ at the $10^{-3}$ level. A smooth prior says the next decade tightens the posterior toward a centroid. A computability filter says the value should sit at a preferred discrete point tied to a low-complexity inflationary potential. CMB-S4 and LiteBIRD can discriminate.
Coupling-constant correlations. A smooth prior says deviations of $\alpha_{\text{em}}$, $\alpha_s$, and lepton mass ratios from naive predictions are uncorrelated. A computability filter says they should be correlated. A single short halting certificate constrains many constants simultaneously. This is the strongest signature because correlation tests are statistically powerful.
CMB spectral distortions. If thermalisation is a halt condition rather than a smooth limit, cosmologies near the boundary of the computable subset should show incomplete thermalisation. The $\mu$ and $y$ distortion parameters measure exactly this. A smooth prior predicts no preferred scale beyond ordinary dissipation physics. A computability filter predicts a characteristic scale tied to halting structure. PIXIE-class missions can hit the range.
There is also a negative signature. Statistical priors predict observable tails like rare coincidences at calculable rates. Computability filters are sharp at the boundary of the decidable set. Persistent absence of statistical anomalies that anthropic priors predict is itself evidence for a sharp filter.
References
Cubitt, T., Perez-Garcia, D., Wolf, M. (2015). "Undecidability of the spectral gap." Nature 528, 207-211.
Faizal, M., Krauss, L. M., Shabir, A., Marino, F. (2025). "Consequences of Undecidability in Physics on the Theory of Everything." Journal of Holography Applications in Physics 5(2), 10-21. DOI: 10.22128/jhap.2025.1024.1118.
Faizal, M., Shabir, A., Khan, A. K. (2025). "Consequences of Goedel theorems on third quantized theories like string field theory and group field theory." Nuclear Physics B 1010, 116774.
Shiraishi, N., Matsumoto, K. (2021). "Undecidability in quantum thermalization." Nature Communications 12, 5084.
Strominger, A., Vafa, C. (1996). "Microscopic origin of the Bekenstein-Hawking entropy." Phys. Lett. B 379, 99-104.
Blumberg, M. (2026). "Undecidability Does Not Kill Simulation." https://www.svgn.io/p/undecidability-does-not-kill-simulation
Woit, P. (2025). "String Theory is Not the Theory of the Real World." Not Even Wrong. https://www.math.columbia.edu/~woit/wordpress/?p=14059