i tried to select a string vacuum from the new de sitter holography. here's where it broke.

this is a record of a thing i tried that doesn't work. i'm posting it anyway, because the place it fails is more interesting than the place it starts, and because i think writing up honest dead ends is underrated. nothing here is a claim about reality. it's a claim about what happens when you push three recent ideas as hard as they'll go.

the itch

string theory has a number problem. when you compactify the extra dimensions - fold up the six or seven you can't see - you get to choose how. each choice of fluxes, shapes, and wrappings gives a different low-energy universe with a different vacuum energy. the number of consistent choices is often quoted as around $10^{500}$. this is the landscape.

the embarrassing part isn't that the number is large. it's that we have no principle telling us which one is ours. the usual move is the anthropic principle: of all the vacua, we observe one compatible with life, so naturally we find ourselves in a habitable one. that's not wrong, but it's not satisfying either - it's a filter, not a selector.

i wanted a non-anthropic selector. something that says: out of $10^{500}$, physics itself prefers this one, for structural reasons, and as a bonus spits out the cosmological constant we measure:

that number - the vacuum energy of our universe, in units where the Planck scale is 1 - is the single most precisely unexplained small number in physics. if a selection principle is any good, it should land near it.

why this even looked possible in 2026

for decades the blocker was specific: we didn't know how to do holography for de Sitter space. holography is the idea that a theory of gravity in some region is secretly equivalent to a theory without gravity living on the boundary - the famous example being AdS/CFT. the reason it matters for vacuum selection: a boundary theory gives you a consistency filter. not every gravitational setup has a sensible boundary dual, so demanding that a vacuum admits a good holographic description prunes the landscape.

the catch: our universe isn't anti-de Sitter. it's de Sitter - positively curved, accelerating, $\Lambda > 0$. and dS holography was a swamp. between 2023 and 2025, three pieces showed up that individually chip at this: DSSYK, conjectured dual to three-dimensional de Sitter; the T² deformation, which pushes a 3D CFT toward a dS₄ description; and the CLPW algebra, which makes the de Sitter static patch Hilbert space effectively finite-dimensional, with dimension $\sim e^{A/4G_N}$.

the thought that nagged at me: these three things might be the same chain seen from three angles. so i tried to build it.

link 1 - one number controls everything (in 3D)

start with DSSYK. its single coupling is the chord parameter $\lambda = p^2 / 2N$, where $p$ is the order of the fermion interaction and $N$ is the number of fermions. the proposed dictionary relates this to the de Sitter radius:

in three dimensions the cosmological constant is set by the radius, $\Lambda_3 \sim 1/R_{\text{dS}}^2$. carrying the factor the dictionary suggests:

read that slowly, because it's the whole hook: the entire cosmological constant is a function of one number, $\lambda$. vacuum selection has been reduced to a single question - what sets $\lambda$? hold onto that exponent $\lambda^2$. it does not survive contact with link 5, and the inconsistency is mine to own.

link 2 - the throat makes it tiny, and a number falls out

why is $\Lambda$ so absurdly small? in string compactifications the standard mechanism is a warped throat - a Klebanov–Strassler geometry where a region of space is exponentially redshifted, in the style of KKLT. the coupling inherits the warp factor:

where $g_s$ is the string coupling and $K, M$ are integer fluxes threading the throat. square it and feed it into the boxed formula:

now the satisfying moment. set $G_N = 1$, pick generic small-coupling values $g_s = 0.1$, $N_D = 1$, $M = 1$, demand $\Lambda_3 = \Lambda_{\text{obs}}$, and solve for $K$:

a flux integer comes out near 3. round it: $K = 3$. that's the rush. a messy $10^{-122}$ became a small integer through nothing but an exponential and two generic inputs. for about a day i thought i had something.

you should already feel the first crack: i didn't predict $\Lambda_{\text{obs}}$. i plugged it in and solved backwards for $K$. that's tuning, not derivation. the exponential just makes the tuning land on a clean integer instead of an ugly real. file that. it gets worse.

link 3 - the step where it all falls apart

DSSYK is three-dimensional. we live in four. the single most important thing the chain needed was: does $K=3$ survive the jump to 4D? this was supposed to be the keystone. here is exactly how it dies.

the T² deformation gives a 4D coupling. following the dimensional logic, after reducing the boundary CFT₃ on an $S^2$ down to an effective DSSYK quantum mechanics, the chord coupling becomes $\lambda_{\text{4D}} = \frac{G_N}{4\pi}\sqrt{\Lambda/3}$, using $\Lambda = 3/\ell_{\text{dS}}^2$. and the 4D analogue of the link-1 formula is $\Lambda_{\text{4D}} = 48\pi^2 \lambda_{\text{4D}}^2 / G_N^2$. substitute one into the other:

i got $\Lambda = \Lambda$. that's it. that's the keystone. it's an identity - a tautology dressed up in $\pi$'s. i defined $\lambda_{\text{4D}}$ in terms of $\Lambda$, then plugged it into a formula for $\Lambda$, and unsurprisingly recovered the $\Lambda$ i started with. the flux $K$, the integers $M$, the coupling $g_s$ - none of them appear anywhere in this step. there is no $K$-dependence to survive, because $K$ never entered.

the claim i most wanted to make - $K=3$ is confirmed in 4D - is not under-proven. it is un-stated. the math that was supposed to contain it contains nothing.

the reckoning

once link 3 fell, i went back and looked at the rest without the rose tint. $\Lambda$ is inserted by hand, twice. in link 2 i solved backwards from $\Lambda_{\text{obs}}$ to get $K$. later i'd set the DSSYK fermion count $N = S_{\text{dS}} = 10^{122}$ - but $S_{\text{dS}}$ is itself just $\sim 1/\Lambda$. i was feeding the answer in at two separate points and then acting surprised when the answer came out.

the exponent isn't even consistent. link 1 says $\Lambda \propto \lambda^2$. but later, taking $\lambda \approx 4.5\times10^{-122}$ and reporting $\Lambda \approx 10^{-121}$, i was implicitly using $\Lambda \propto \lambda$. with the quadratic law that same $\lambda$ would give $\Lambda \sim 10^{-243}$, off by 120 orders of magnitude. i never reconciled them.

the selection constraints fail at my own chosen point. supergravity validity wants $g_s M \gg 1$; mine is $0.1$. tadpole cancellation wants $K M \sim 10\text{–}20$; mine is $3$. two of three conditions are violated by the very numbers i chose. and the bridges between frameworks - the M-theory hyperbolic compactification mapped to the KKLT throat - were asserted, not derived.

and the foundation is contested. whether DSSYK is really dual to dS₃ is an open conjecture. the de Sitter swampland program argues that controlled, metastable de Sitter vacua might not exist in string theory at all. i built a selection principle on top of scaffolding the field hasn't agreed is load-bearing.

what's actually left standing

strip away everything i assumed, and one question survives that is genuinely well-posed and checkable:

does the singlet sector of ABJM theory on $S^2 \times \mathbb{R}$, under the T² deformation, double-scale to DSSYK with chord order $p$ equal to the Chern-Simons level $k$?

that's not a grand claim. it's a concrete technical question with a yes-or-no answer that someone who knows the representation theory could settle. if yes, there's a small real result buried here. if no, the chain has no spine and that's worth knowing too.

what i actually learned

an exponential will always hand you a clean integer if you solve backwards through it. $K \approx 3$ felt like discovery. it was arithmetic. any time a tiny number turns into a tidy one via $\exp(\text{stuff})$, check whether you put the tiny number in by hand first. i had.

the step you're proudest of is the one to attack hardest. i wanted link 3 to be the keystone so badly that i didn't actually run the cancellation until late. the moment i did, $\Lambda = \Lambda$ stared back. if i'd done the algebra on day one instead of the narrative on day one, i'd have saved a week.

a clean dead end is a real result. i can now say precisely why this particular assembly of these particular 2023–2025 ideas does not select a vacuum, and i can point at the one sub-question that's worth a real expert's time. that's worth more than a confident wrong paper, and it's the honest state of where i got to.