Selective Souslin-Tree Erasure and the HOD Hypothesis

Set theorists can build universes in which a cardinal is enormous in the full universe yet looks small inside HOD. HOD is the inner model of sets definable from ordinals, roughly the part of the universe that can be "named" without arbitrary choices. One known way to do this is to plant a Souslin tree, a very thin, very tall branching structure with no long path through it and no large family of mutually incomparable nodes, so that HOD can see the tree but cannot see any path through it. This kills the cardinal's "tree property" inside HOD and with it every large-cardinal property that depends on that property. Woodin's HOD Hypothesis, a conjecture about how closely HOD tracks the full universe, points the other way. If it holds and a supercompact cardinal exists, HOD must contain many measurable cardinals. These two facts pull against each other, and nobody has located the line between them.

This note proposes to find that line. It asks how much large-cardinal structure can be erased from HOD with Souslin trees while the HOD Hypothesis still holds. It also tests the erasure method at the low end of the large-cardinal scale, using remarkable cardinals, whose indestructibility under forcing has only recently been worked out. The deliverable is a set of consistency theorems, each with a matching obstruction where we can prove one. Of the three directions in the brief, the first two supply the mechanism and the question. The third, bimodular elasticity, has no mathematical content we can use honestly here; the last section says so directly instead of inventing an analogy.

Core Claim

Two published results fix the frame. Large cardinals need not be large in HOD forces a weakly homogeneous $\kappa$-Souslin tree $T$, codes the universe so that $T$ becomes ordinal-definable, and arranges for HOD to look like $V[T]$. The result is that $\kappa$ stays measurable or supercompact in $V$ but is not weakly compact in HOD. Its class version (Theorems 8 and 9) produces universes with a proper class of supercompacts and no measurables at all in HOD. The HOD Hypothesis and a supercompact cardinal proves in its Theorem 4.3 that a supercompact $\kappa$ plus the HOD Hypothesis yields HOD-measurable cardinals cofinally below $\kappa$.

Putting these together gives a fact neither paper states. In the Theorem 9 model, the HOD Hypothesis must fail, because Theorem 4.3 would otherwise supply HOD-measurables. So total erasure by Souslin trees refutes the HOD Hypothesis.

Our claim is that selective erasure does not. We aim to prove the following. If ZFC plus a supercompact cardinal is consistent, so is a universe in which the HOD Hypothesis holds, $\kappa$ is supercompact, and every cardinal below $\kappa$ that is measurable in $V$ fails to be weakly compact in HOD. In such a universe, the HOD-measurables that Theorem 4.3 forces into existence must be cardinals that are not measurable in $V$. Reflection into HOD would then be a statement about some large cardinals existing in HOD, not about the particular ones present in $V$ surviving there. If instead an obstruction appears, for example if erasing cofinally many $V$-measurables below $\kappa$ provably makes some regular cardinal $\omega$-strongly measurable in HOD, that is equally informative. It would be a positive transfer theorem showing that the HOD Hypothesis protects particular large cardinals.

Research Program

The first stage calibrates the method at the remarkable level. Indestructibility properties of remarkable cardinals builds remarkable Laver functions. It makes remarkability indestructible under ${<}\kappa$-closed, ${\le}\kappa$-distributive forcing and under two-step forcings $\operatorname{Add}(\kappa,\theta) * \dot R$, and it shows that the canonical GCH forcing preserves remarkability. These are the ingredients the HOD-erasure template needs.

The construction runs as follows. Adding one Cohen subset of $\kappa$ is equivalent to first adding a homogeneous $\kappa$-Souslin tree $T$ and then a path $b$ through it. Coding above $\kappa$ by a GCH pattern is ${\le}\kappa$-distributive, so the preparation shields it. The coding makes $T$ ordinal-definable but leaves $b$ undefinable. Remarkable cardinals are weakly compact, so $\kappa$ is then not remarkable in HOD while it stays remarkable in $V$. The point of this stage is not a headline result. It is to check exactly which steps of the coding-compatibility argument in Large cardinals need not be large in HOD (Theorem 5) need the lifting criterion for virtual embeddings, as opposed to real ones.

The class version is the real test at this level: every remarkable cardinal survives in $V$ and none is weakly compact in HOD. It needs an Easton-support iteration in which remarkable Laver functions at different cardinals do not interfere. That is where virtual large cardinals may behave differently from supercompacts.

The second stage is selective erasure below a supercompact. We start from a model of V=HOD with a Laver-indestructible supercompact $\kappa$; V=HOD makes the HOD Hypothesis hold trivially. We then iterate the tree-plus-coding step at every $V$-measurable $\gamma<\kappa$, and possibly at every remarkable one. At $\kappa$ itself and on a reserved class of regular cardinals we do nothing, so stationary subsets of their cofinality-$\omega$ points can still be split into pieces inside HOD.

The technical heart is computing the HOD of the final model. The coding must put enough of the ground model into HOD to carry the splittings that witness the HOD Hypothesis. It must not put in the paths $b_\gamma$, because those would restore weak compactness. Weak homogeneity of the remaining forcing is what keeps them out. Preservation of supercompactness through the iteration follows the directed-closure pattern used in Souslin trees and successors of singular cardinals, but the measurables being erased are below $\kappa$, so Laver functions guide the lifts.

The third stage looks for the obstruction. We run the argument of Theorem 4.3 inside the stage-two model, track which cardinals it produces as HOD-measurable, and ask whether any proof of the HOD Hypothesis can avoid them. In parallel we try to prove the dual statement. That statement says the following: if Souslin trees are added homogeneously at cofinally many $\gamma<\kappa$, then every limit of such $\gamma$ of cofinality $\omega$ produces an $\omega$-strongly measurable cardinal in HOD. Whichever side closes settles the question.

On bimodular materials: we checked whether that direction could supply a method rather than a metaphor. The only structural echo is that Equilibrium Configurations for Nonhomogeneous Linearly Elastic Materials with Surface Discontinuities faces a compactness failure in which pieces of the material drift away and must be corrected. That loosely resembles a tree with no path. The resemblance does no logical work, so that direction is treated only as a scope constraint. Nothing in the proposal depends on it.

Baselines

The do-nothing reference point is the existing literature taken together. It already gives erasure of a single supercompact or measurable from HOD, total erasure with the HOD Hypothesis therefore failing, and HOD-measurables guaranteed under the HOD Hypothesis. The proposal beats this baseline only if it produces a model where selective erasure and the HOD Hypothesis coexist, or a theorem forbidding that. Restating the tension does not count.

The strongest existing method is the coding-compatibility template, Theorem 5 of Large cardinals need not be large in HOD. To compare fairly, we apply it with the same large-cardinal hypothesis, a single supercompact, and the same preparatory forcing that we use. We then record which of our target properties it delivers. Out of the box it erases at a chosen $\kappa$, and its class version erases everywhere. Neither says anything about the HOD Hypothesis. Beating it means proving the HOD Hypothesis in the final model, which the template never attempts.

At the remarkable level, the baseline is the indestructibility theorem of Indestructibility properties of remarkable cardinals applied as-is. Our class-erasure theorem beats it only if it handles simultaneous Laver functions, which that paper does not.

The key-ingredient ablation runs the stage-two iteration without the reserved class of untouched regular cardinals, erasing at every measurable including $\kappa$. By the observation above, this ablation should make the HOD Hypothesis fail. If it does not fail, our account of where the HOD Hypothesis comes from is wrong.

These are consistency results, not measurements, so there is no metric to report. No reported result found on Lacuna for any quantitative comparison in Directions 1 or 2.

Why not just…?

Why not just apply the coding-compatibility theorem to remarkable cardinals and stop? This is a real threat to stage one. Remarkability may simply turn out to be coding compatible using the indestructibility results already published. That is why stage one is presented as calibration rather than contribution. The proposal's weight rests on the class version and on stage two, which no existing template addresses.

Why not just put remarkable cardinals into the singular-successor Souslin tree construction? The forcing $P_\kappa$ of Souslin trees and successors of singular cardinals is $\kappa$-directed closed and adds no new sets of size below $\lambda^+$. That makes it ${<}\kappa$-closed and ${\le}\kappa$-distributive, so the remarkable indestructibility theorem already preserves remarkability under it. We expect this to be a near-immediate corollary, not a project, and we do not claim it.

Why not just erase everything, as in Theorem 9, and study that model? That model already answers the question in the negative direction: combined with Theorem 4.3 of The HOD Hypothesis and a supercompact cardinal, it forces the HOD Hypothesis to fail. The open content lies strictly between total erasure and none, which is exactly what selective erasure probes.

Why not just build a bridge to bimodular mechanics, for example reading neutral-layer migration as branching, or using physics-informed or symbolic PDE solvers as a test bed? Tools like Coupled Integral PINN for Discontinuity and the benchmark PINNacle work with real-valued fields on bounded domains. Nothing in them can express a forcing extension or an uncountable tree, so such a bridge would be metaphor. We decline it explicitly.

Decisive Checks

The first check is the ablation prediction. Full erasure including $\kappa$ must refute the HOD Hypothesis by Theorem 4.3. If our HOD computation says otherwise, the computation is wrong. This catches errors before the main theorem rests on them.

The second check, in the selective model, is to name the cardinals that Theorem 4.3 makes HOD-measurable and verify they are not measurable in $V$. If they always turn out to be $V$-measurable, that points toward the obstruction theorem.

The third check is to confirm in HOD the stationary-set splittings at the reserved regular cardinals. These splittings are the direct witness for the HOD Hypothesis.

Success means either a model with selective erasure in which the HOD Hypothesis holds, or a proof that erasing cofinally many measurables below a supercompact produces $\omega$-strongly measurable cardinals in HOD. Either outcome sharpens the conjectured failure of the converse to Theorem 4.3 discussed in The HOD Hypothesis and a supercompact cardinal.