Geometric Cobordism Classes for Non-$\mathbb{A}^1$-Invariant Motivic Spectra over Arbitrary Rings

Algebraic cobordism is the algebraic analogue of the topologist's bordism of manifolds. It is the universal way to attach characteristic classes, such as Chern classes of vector bundles, to spaces built from polynomial equations. There are now two quite different versions of it.

The abstract version, $\mathrm{MGL}$, is due to Annala, Hoyois and Iwasa. It lives in a homotopy theory that no longer treats the affine line as contractible ("non-$\mathbb{A}^1$-invariant"), and it is defined over any quasi-compact quasi-separated base, including arbitrary commutative rings. The geometric version is Annala's bivariant derived cobordism. It is built from explicit cycles: maps of derived schemes, optionally carrying a vector bundle, modulo explicit relations. Its classes can be written down and computed. But it has only been constructed under restrictive hypotheses. Annala and Yokura need a Noetherian base of finite Krull dimension. The base-independent theory needs an ample family of line bundles. The Spivak-type generation results need a field or a nice discrete valuation ring (DVR) with the residual characteristic inverted.

This note proposes to close that gap. The aim is to build the geometric theory with bundles over every commutative ring, and to compare it with the abstract $\mathrm{MGL}$ through an explicit map.

The core claim has two parts. First, removing the Noetherian hypothesis is largely a formal problem, handled by writing a ring as a filtered colimit of finitely generated subrings. We intend to prove this rather than argue around it. Second, the real mathematical content is the comparison map from geometric cycles to abstract $\mathrm{MGL}$ classes. Continuity lets us study that map base by base and test it on rings where Noetherian intuition fails: valuation rings of rank at least two, perfect $\mathbb{F}_p$-algebras, and infinitely generated non-reduced rings.

Where the literature stands

Four papers supply the method; several others motivate or constrain it.

Algebraic cobordism and a Conner–Floyd isomorphism for algebraic K-theory does three things over any qcqs derived scheme. It replaces $\mathbb{A}^1$-invariance by $\mathbb{P}^1$-invariance plus a weighted form of $\mathbb{A}^1$-invariance. It constructs $\mathrm{MGL}$ as the initial oriented ring spectrum. It proves $\mathrm{MGL}^{**}(X)\otimes_{\mathbb{L}}\mathbb{Z}[\beta^{\pm1}]\simeq K^{**}(X)$, where $\mathbb{L}$ is the Lazard ring (the ring classifying formal group laws) and $\beta$ is the Bott element. This paper already makes the abstract theory non-Noetherian, so "extend motivic spectra to non-Noetherian rings" is not an open problem and we do not propose it. What that paper leaves open is a direct, integral comparison with geometrically defined cobordism groups.

Motivic spectra and universality of $K$-theory proves that $K$-theory is the universal theory satisfying the projective bundle formula generated by line bundles. Its constructions are Zariski-local, and it notes that transfers are not constructed in general. Cohomology of the moduli stack of algebraic vector bundles computes $E^*(\mathscr{V}\mathrm{ect}_n)$ as a power series ring in Chern classes, assuming $E$ has finite quasi-smooth transfers. That assumption is exactly what a geometric cycle model would supply, since pushforward of a cycle is a transfer by construction.

On the geometric side, three papers supply the construction:

  • Derived Algebraic Cobordism builds bivariant cobordism from derived blowups. It proves a projective bundle formula and recovers $K^0$ via the multiplicative formal group law. It leaves the comparison with abstract $\mathrm{MGL}$ open.
  • Bivariant algebraic cobordism with bundles adds bundles to cycles, $[V\to X, E]$. It proves only a weak projective bundle formula, and only over Noetherian bases of finite Krull dimension.
  • Base independent algebraic cobordism shows the groups do not depend on the base ring, and extends to divisorial derived schemes. Its Chern character and Riemann–Roch arguments use the resolution property, and it leaves open (Question 3.32) whether $\Omega^\bullet$ agrees with its variant $\Omega'^\bullet$, the variant in which Euler classes are nilpotent.

This base-independence result is the single most important input for us. It says the base ring is almost irrelevant, which is what makes a colimit argument plausible.

Two further lines of work constrain the design. The Noetherian presentation and Gersten machinery in Gabber's presentation lemma over noetherian domains and A Nisnevich local Bloch-Ogus theorem over a general base depends on dimension induction, infinite residue fields and $\mathbb{A}^1$-locality. We therefore deliberately avoid any strategy that routes through Gersten complexes. Reductions of Ideals in Commutative Rings shows that ideal-theoretic structure need not pass to quotients in non-Noetherian rings. This warns us that any argument touching non-finitely-generated ideals will not survive approximation. Our constructions must only ever touch finitely presented data.

Core Claim

Let $R=\varinjlim R_i$ be the filtered colimit of its finitely generated $\mathbb{Z}$-subalgebras. We conjecture, and propose to prove, three statements.

First, the bivariant theory with bundles, restricted to finitely presented quasi-smooth morphisms, is continuous: $$\Omega^*(X\to Y)\;\cong\;\varinjlim_i \Omega^*(X_i\to Y_i).$$ Here $X_i\to Y_i$ is any finitely presented model over $R_i$ of the morphism $X\to Y$ over $R$. The heuristic reason is that generators and bordism relations are both finitely presented data, so each descends to some $R_i$. We also expect the abstract groups $\mathrm{MGL}^{2n,n}$ of Annala–Hoyois–Iwasa to be continuous in the same sense.

Second, the strong projective bundle formula for arbitrary bundles $\mathbb{P}(E)$, and the full theory of Chern classes of higher-rank bundles, hold over every commutative ring. Over non-Noetherian rings neither is currently available for the geometric theory with bundles.

Third, there is a natural map $\Phi:\Omega^n(X)\to \mathrm{MGL}^{2n,n}(X)$, obtained by sending a cycle to the pushforward of its fundamental class. After tensoring both sides with $\mathbb{Z}[\beta^{\pm1}]$, this map becomes an isomorphism onto $K_0(X)$ with integer coefficients, for all qcqs $X$ with the resolution property. This would give a geometric, integral Conner–Floyd statement at the level of groups.

The first two statements are the engine. The third is the payoff, and it is where genuinely new input is needed. Inside the non-$\mathbb{A}^1$-invariant category, $\Phi$ requires constructing quasi-smooth pushforwards on $\mathrm{MGL}$. That is precisely the transfer problem flagged as open in Motivic spectra and universality of $K$-theory.

Research Program

Step 1: continuity. The first step is purely about finiteness. We check four things:

  • every relation used to define bivariant cobordism with bundles is witnessed by a finitely presented derived scheme over some $R_i$;
  • quasi-smoothness and projectivity descend along filtered colimits;
  • the weak projective bundle formula of Annala–Yokura is compatible with base change $R_i\to R_j$;
  • the base-independence isomorphism of Annala makes the colimit independent of the chosen presentation.

On the abstract side, continuity of $\mathrm{MGL}^{**}$ should follow from compactness of the relevant objects in the category of Annala–Hoyois–Iwasa. We will check that the weighted $\mathbb{A}^1$-localization does not break compactness.

Step 2: strong projective bundle formula. Over a Noetherian finite-dimensional base, this should follow from the weak formula plus Zariski-local trivialization of $E$ and a Mayer–Vietoris argument. Continuity then transports it to any ring. Chern classes of higher-rank bundles follow by the Grothendieck method.

Step 3: construct $\Phi$ and prove the Conner–Floyd statement. We construct $\Phi$ by showing that $\mathrm{MGL}$ admits pushforwards along projective quasi-smooth maps. The route is to produce Thom classes from the orientation and use the Pontryagin–Thom factorization through a projective bundle, which Step 2 now provides. We then prove the integral Conner–Floyd statement for $\Omega$ geometrically. Derived Algebraic Cobordism already recovers $K^0$ from the multiplicative formal group law. Combined with continuity, this removes all characteristic and dimension hypotheses.

Step 4: settle what approximation cannot. Over a rank-two valuation ring $V$, the group $\Omega^*(\mathrm{Spec}\,V)$ should be forced by continuity. Questions about all coherent sheaves, such as the resolution property and Question 3.32, are not about finitely presented data. Here we expect genuine failures, and we will say exactly where approximation stops working.

Baselines

This is a theorem-proving program, not an empirical one. The baselines are the strongest existing results that answer the same question, and "beating" one means proving a statement it cannot reach.

The naive reference point is to define $\Omega^*$ over an arbitrary ring $R$ as the colimit over Noetherian subrings, with no further theorems. It is honest but empty. It says nothing about projective bundle formulas, Chern classes or comparison with $\mathrm{MGL}$ over $R$. We beat it if Steps 2–3 hold for the colimit-defined object, and in particular if the colimit agrees with the intrinsic cycle definition over $R$.

The strongest existing competitor for the comparison question is Algebraic cobordism and a Conner–Floyd isomorphism for algebraic K-theory. It already works over all qcqs bases and already relates $\mathrm{MGL}$ to $K$-theory. We compare on the same bases and the same class of schemes. The proposal adds value only if $\Phi$ exists, is compatible with their Conner–Floyd map, and yields an integral isomorphism of geometric groups. That integral statement is exactly what their paper leaves open. No reported numerical result on Lacuna exists for this comparison; the relevant outcomes are theorems.

The strongest geometric competitor is Bivariant algebraic cobordism with bundles, restricted to Noetherian bases of finite Krull dimension. We beat it by removing that restriction and upgrading its weak projective bundle formula to the strong one.

The ablation that isolates the key ingredient is the bundle-free theory of Base independent algebraic cobordism. If the Conner–Floyd statement for $\Phi$ goes through without bundles on cycles, then the bundles are not doing work for this question. We would then report the simpler statement, and the bundle theory would matter only for rank-$r$ refinements.

The benchmarks in the research file do not measure anything relevant here. NaturalProofs, for example, evaluates retrieval over informal proofs. We do not use them, and no reported result was found on Lacuna for any computation of cobordism groups over non-Noetherian rings.

Why not just…?

Why not just use Noetherian approximation and stop? This is the real threat, and the proposal adopts it rather than competing with it. Step 1 is Noetherian approximation. The claim of novelty rests entirely on Steps 3–4. Step 3 is the construction of $\Phi$, which requires transfers on $\mathrm{MGL}$ that approximation does not provide. Step 4 is the precise identification of where approximation fails: statements about non-finitely presented sheaves, such as the resolution property that Base independent algebraic cobordism relies on. If Step 1 turns out to give everything, including $\Phi$, we will say so and report the work as a clean formal reduction.

Why not just restrict to coherent rings or rank-one valuation rings? Continuity makes such restrictions unnecessary for finitely presented questions. For the other questions, the interesting failures occur exactly at rank at least two and in infinite Krull dimension. Algebraic Spivak’s theorem andapplications already shows that even nice DVRs force inverting the residual characteristic for generation results. Restricting further would not remove that barrier.

Why not just work with abstract $\mathrm{MGL}$ and skip geometric cycles? Abstract $\mathrm{MGL}$ gives no explicit classes and no transfers in the non-$\mathbb{A}^1$-invariant setting. The Chern class computation in Cohomology of the moduli stack of algebraic vector bundles needs those transfers. Step 3 would show whether geometric cycles supply them. If $\mathrm{MGL}$ turns out to admit transfers by purely formal means, the geometric model is still useful for computation, but its structural role shrinks.

Why not just impose $\mathbb{A}^1$-invariance and reuse the Gersten machinery? That discards $K$-theory of singular schemes, which is the target of Conner–Floyd. The Gersten route also needs Noetherian, regular, J-2 bases with infinite residue fields, as in A Nisnevich local Bloch-Ogus theorem over a general base.

Decisive Checks

We will run five explicit test computations, each of which can confirm or refute the program.

  1. Dual numbers in infinitely many variables. For $R=k[x_1,x_2,\dots]/(x_ix_j)$, verify that $\Omega^*(\mathrm{Spec}\,R)$, computed by continuity, maps under $\Phi$ and $\beta$-inversion isomorphically onto $K_0(R)$. This tests integral Conner–Floyd on a non-reduced, non-Noetherian ring where $\mathbb{A}^1$-invariant theories see nothing.
  2. Rank-two valuation ring. For a rank-two valuation ring $V$, compute $\Omega^*(\mathbb{P}^n_V)$ and $\Omega^*(\mathbb{P}(E))$ for a nontrivial bundle $E$. Check both against the strong projective bundle formula.
  3. Perfect $\mathbb{F}_p$-algebras. For a perfect ring such as $\mathbb{F}_p[t^{1/p^\infty}]$, test whether the low-degree vanishing of Spivak type survives without inverting $p$. This probes the integrality barrier directly.
  4. Rank-two Chern classes. Test whether the classes $c_1,c_2$ of a rank-two bundle over such bases satisfy the Whitney formula compatibly with the power series description of $E^*(\mathscr{V}\mathrm{ect}_2)$.
  5. Question 3.32 over a non-Noetherian base. Attempt to settle whether $\Omega^\bullet$ and $\Omega'^\bullet$ agree over a non-Noetherian base. A counterexample there would itself be a publishable boundary result.

A negative outcome in the first check, meaning $\Phi\otimes\mathbb{Z}[\beta^{\pm1}]$ fails to be injective integrally, would refute the core claim.