Known theorems · a proposed dynamical picture
Foliations, singular points, and Deninger’s picture of the zeros
A foliation cuts a space into thinner pieces, called leaves. At a singular point a leaf can shrink to a zero of a vector field. Christopher Deninger has proposed that the zeros of ζ might appear as eigenvalues of a flow on a foliated space, with closed orbits playing the role of primes. The space that would make this a theorem about ζ has not been constructed. Nothing on this page is progress on the Riemann Hypothesis.
[known] How to read the labels. Every claim carries one tag. [known] is published mathematics, with the source named (full list in the sources and in the primer). [project] is a figure or check we made ourselves; it illustrates a known fact, it is not a new theorem. [speculative] is a proposed picture that has not been constructed, or a reading of one. Click any tag to see every claim that carries it.
Nothing on this page is progress on the Riemann Hypothesis. A dictionary that would, if a certain space existed, recast RH as a positivity or Hodge-type condition is not a proof, and is not evidence.
This page is a first illustrated primer. The longer write-up, with the same labels, is primer.md. Related pages: the explicit formula and the honest status of RH.
What is a foliation?
[known] Imagine a manifold — a space that looks locally like ordinary Euclidean space of some dimension n — cut into a family of thinner pieces, all of the same dimension p, that fit together smoothly and never cross. Each piece is called a leaf. A foliation of dimension p (equivalently, of codimension n − p) is precisely such a partition, with a compatibility condition in charts: around every point one can find coordinates in which the leaves look like a stack of parallel plaques, the level sets of the remaining coordinates. Changing charts may shuffle the plaques, but plaques go to plaques.
The leaves need not be compact, and they need not sit as closed subsets. A single leaf may wind densely through the whole space. That already happens on a torus.
Standard references: Candel–Conlon, Foliations I; Camacho–Lins Neto, Geometric Theory of Foliations; Haefliger’s 1958 paper on foliated structures.
Example: the Kronecker foliation of the torus
[known] Let T2 be the two-torus, drawn as the unit square with opposite sides glued. Fix a slope α, and take the lines of that slope in the plane, then descend them to the torus. If α is rational, every leaf is a closed circle. If α is irrational, every leaf is dense: a single line comes arbitrarily close to every point, infinitely often, and never closes. The animation below lets you switch between the two cases and watch a point travel along one leaf.
Figure 1
Example: the Reeb foliation of the solid torus
[known] The Reeb foliation of the solid torus D2 × S1 (Reeb, 1952) has a single compact leaf — the boundary torus — and every interior leaf is a plane, winding and accumulating on that boundary. In a meridional disk (a slice of the solid torus), the traces of the interior leaves appear as curves that pile up against the boundary circle. Gluing two such Reeb solid tori along their common boundary produces the Reeb foliation of the three-sphere S3.
Figure 2
Example: foliations by level sets
[known] If f is a smooth real-valued function whose derivative never vanishes, the level sets f−1(c) are a foliation of codimension one. If the derivative is allowed to vanish, those zeros are singular points of the would-be foliation: the regular levels no longer cover the whole space, and near a critical point the levels can change shape. A first picture is f(x, y) = x2 + y2 on the plane: away from the origin the levels are circles, and the origin is a singular fibre.
Singular points, and zeros of a vector field
[known] A regular foliation has a leaf through every point, all of the same dimension. Many naturally occurring “foliations” are allowed to have singular points, where a defining vector field or holomorphic form vanishes. This is the precise meaning of a foliation with a zero point.
If Y is a smooth vector field on a surface and p is an isolated zero, the orbits of Y (except p itself) are the leaves of a singular foliation, with a singularity at p. When the linearisation at p is non-degenerate, the local picture is one of four linear types, distinguished by the eigenvalues of the linearisation. The Poincaré–Hopf index of the zero is the winding number, around a small circle about p, of the normalised field Y/|Y|.
Figure 3
Holomorphic foliations: Seidenberg reduction and the Camacho–Sad index
[known] In complex dimension two the natural object is a holomorphic foliation, given locally by a holomorphic vector field, or by a holomorphic 1-form ω = A dx + B dy. A singularity is a point at which both A and B vanish.
Seidenberg’s reduction of singularities (1968). Starting from an isolated singularity of a holomorphic foliation of a complex surface, a finite sequence of blow-ups produces a foliation whose only singularities are reduced: in local coordinates the linearisation has at least one nonzero eigenvalue, and if both are nonzero their ratio is not a positive integer. The precise definition is in Seidenberg’s paper and in Brunella, Birational Geometry of Foliations, Chapter 1. The point for this primer is that singularities can be simplified, by a canonical geometric process, to a short list of local models.
The Camacho–Sad index (1982). If S is a separatrix through a singularity p — a germ of an invariant analytic curve through p — Camacho and Sad attach a complex number CS(F, S, p). If C is a compact invariant curve, the sum of the Camacho–Sad indices at the singular points of C equals the self-intersection C · C. In particular, every isolated singularity of a holomorphic foliation of a surface admits at least one separatrix.
These are the standard local invariants of a singular holomorphic foliation. They are recorded here because “a foliation with a zero point” is, in this category, a rich and completely rigorous subject, independent of any arithmetic dictionary.
Deninger’s proposed dictionary
[known] as a proposal. In a series of papers beginning with a 1992/1994 motives paper and set out for a general audience in the 1998 ICM address and a 2008 survey, Christopher Deninger has proposed to compare an arithmetic scheme X over Spec Z with a flow φt on a foliated space (X, F). Closed orbits of the flow would play the role of primes, with length ℓ(γ) corresponding to log p. A dynamical Lefschetz, or transversal index, trace formula would match the explicit formula of analytic number theory. The zeros of a zeta function would appear as eigenvalues of the infinitesimal generator θ of the flow, acting on reduced leafwise cohomology.
Deninger is explicit, in the first paragraph of the 2008 survey, that the cohomology theory he wants for schemes over Spec Z “has not been constructed.” What has been constructed is a class of dynamical systems whose leafwise cohomology has several of the formal properties that such a cohomology would need, and a handful of genuine examples — function fields, solenoids, a circle — in which pieces of the dictionary can be checked. The diagram below is his published dictionary, not an extension of it.
Figure 4
Leafwise cohomology, the trace formula, and what RH would become
[known] On a foliated space one can differentiate only in the leaf direction. The resulting de Rham complex along the leaves has a reduced leafwise cohomology Ḧ•F(X). Álvarez López and Kordyukov proved that, for a Riemannian foliation, this is the kernel of the leafwise Laplacian — a Hodge theorem in the leafwise direction. The result needs the Riemannian hypothesis; Deninger and Singhof gave a counterexample for general foliations.
The flow, because it maps leaves to leaves, acts on this cohomology. Its infinitesimal generator is an operator θ. Deninger’s proposal is that the non-trivial zeros of the zeta function should correspond to the eigenvalues of θ on the degree-one (or middle-degree) reduced leafwise cohomology.
[known] Under strong conformal hypotheses on a compact 3-manifold, a transversal index theorem of Álvarez López–Kordyukov and Deninger–Singhof gives a distributional identity of the same shape as the explicit formula, and every eigenvalue ρ of θ satisfies Re ρ = α/2, where α is the conformal factor of the flow. On manifolds those hypotheses force α = 0 (purely imaginary eigenvalues). To have α = 1, the value that would match Re ρ = 1/2, one must leave the manifold category and work with solenoids; Deninger constructs such examples from elliptic curves over finite fields, with α = 1.
[known] as a proposed translation, not a theorem. Matching the explicit formula for ζ would require α = 1, hence Re ρ = 1/2. In the 1998 ICM address Deninger discusses this as a positivity or Hodge-type statement on the still unconstructed arithmetic cohomology. [speculative] That translation presupposes the missing space. It is not a proof of the Riemann Hypothesis, and not evidence for one.
A related picture, which is not the same
[known] Alain Connes’ 1999 Selecta paper constructs a spectral realization of the zeros of ζ as an absorption spectrum of an operator on an adelic space, in the language of noncommutative geometry, and recasts the Riemann hypothesis as a form of Weil positivity. The two programmes share a family of ideas — a dynamical or spectral reading of the explicit formula, a role for positivity — but they are not the same construction. This page is about Deninger’s foliated picture. Connes is mentioned only to keep the two from being conflated.
Three toy checks you can follow by hand
[project] Each of the three examples below was checked by us on 13 November 2026, using PARI/GP and interval arithmetic (Arb). In every case the thing being checked is a [known] identity from the literature. These are sanity checks of published mathematics on small models. No solenoid has been constructed here, and this is not a claim about the Riemann Hypothesis. The full write-up is compute-20261113.md, with scripts and raw results in circle-toy-results.json, poincare-hopf-results.json and elliptic-orbits-results.json.
The circle toy: a determinant that equals 1 − q−s
[known] Take a circle of length L = log q, and let time act on it by rotation at unit speed. This is the simplest possible “flow on a space,” and it has exactly one closed orbit, of length log q. The generator θ = d/dx of the flow has eigenvalues 2πiν / log q, one for each integer ν. Deninger’s question is: what is the “determinant” of s − θ, that is, the product of all the numbers s − 2πiν / log q? That product is infinite, so it has to be regularized, which means assigning it a finite value by a standard analytic recipe (a zeta-function regularization using the Hurwitz zeta function). The answer, going back to Lerch, is
det∞(s · id − θ) = 1 − q−s.
The zeros of the right side are exactly the numbers s = 2πiν / log q, the eigenvalues of θ. So here the “zeros” are eigenvalues of a generator, in a case where everything can be computed. Source: Deninger, “Analogies between analysis on foliated spaces and arithmetic geometry” (2008), the section called “The simplest example”.
[known] The proof uses three standard facts: ζH(0, a) = ½ − a, ζH′(0, a) = log(Γ(a)/√(2π)), and the reflection formula Γ(z)Γ(1 − z) = π / sin(πz). Put together, they give 2 sin(πα) eiπ(α − 1/2) with α = i s log q / (2π), and that simplifies to 1 − q−s. The write-up fixes the branch of the logarithm for real s > 0 and then extends to all s by analytic continuation, since the right side is an entire function.
[project] As a check of the algebra, we evaluated the Gamma-function expression and 1 − q−s at 384 bits of precision with Arb, on 58 pairs (q, s) with q ∈ {2, 3, 5, 10} and s both real and complex. Every enclosure of the difference contains 0, with typical radius around 10−115. A rough finite-difference check of the derivative g′(0) was loose and is not used as evidence.
Figure 5
[known] The leading coefficient. The Ruelle zeta function of this flow is ζR(s) = 1 / (1 − q−s). Near s = 0 we have 1 − q−s = s log q − …, so ζR has a simple pole at s = 0, and s · ζR(s) tends to 1 / log q. In Deninger’s “Lichtenbaum-style” language, that number is the inverse of a cup-product determinant on the cohomology of the circle, which equals log q, with no torsion. This is an exact calculation, no numerics needed.
[project] The figure below plots s / (1 − q−s) for q = 2, 3, 10 and shows it approaching 1 / log q as s shrinks. The samples differ from 1 / log q by about s / 2, as the expansion predicts, so they are close but correctly nonzero.
Figure 6
Counting zeros of a vector field: three Poincaré–Hopf examples
[known] Poincaré–Hopf says: draw a vector field with finitely many zeros on a closed surface, assign each zero its index (+1 for a source, sink or centre, −1 for a saddle, in the simple cases), and the indices add up to the surface’s Euler characteristic. That number is 2 for a sphere and 0 for a torus. It does not depend on which field you chose, which is why a field with a different zero pattern must still balance. (Hopf 1927; Milnor, Topology from the Differentiable Viewpoint.)
[project] We checked three examples, using the sign of the Jacobian determinant and an independent count of how many times the field turns around a small circle:
- Height gradient on the sphere. Flow “uphill” on S2. The north pole is a sink (+1), the south pole is a source (+1). Sum 2 = χ(S2).
- Sine field on the torus. Y = (sin 2πx, sin 2πy) has four zeros, with indices +1, −1, −1, +1: a source, two saddles and a sink. Sum 0 = χ(T2). The Kronecker field from earlier on this page has no zeros at all, an empty sum equal to 0, which is consistent.
- A second field on the sphere. The function f = z2 + x restricted to S2 has two maxima, one saddle and one minimum, with indices +1, +1, −1, +1. Sum 2 = χ(S2).
Figure 7
Closed orbits and closed points: bookkeeping for an elliptic curve over a finite field
[known] This is the one place where Deninger’s dictionary is carried out in full. Let E be an elliptic curve over a finite field with q elements. A closed point of E is an orbit of the Frobenius map (the map that raises coordinates to the q-th power) on the points of E over the algebraic closure. If an orbit has size n, the matching closed orbit in the dictionary has length n · log q, and its “norm” qn equals the number of elements of the residue field of that point. Counting points over the extension fields gives Nn = qn + 1 − (αn + βn), where α and β are the two Frobenius eigenvalues, and the number Pn of orbits of size n comes out of Nn by Möbius inversion. The zeta function is the Euler product over closed points, and equals the rational function (1 − aT + qT2) / ((1 − T)(1 − qT)) in T = q−s. The Riemann Hypothesis for such curves is a theorem of Weil (Hasse, for elliptic curves). Sources: Deninger 2001 and 2008; Silverman, The Arithmetic of Elliptic Curves, Chapter V.
[project] We tested this bookkeeping on ten curve and field pairs (for example y2 = x3 + 1 over F5, y2 = x3 − x over F17, and the curve 11a1 over F7 and F13). In each, the point counts from the formula agree with PARI’s count over Fqn for n up to 16, and with a direct brute-force count for small n. The Euler product over closed points reproduces the rational function exactly, as a power series with integer coefficients, through T16. A longer run on y2 = x3 + 1 over F5 goes through T24. For that curve the numbers of closed points of degree 1, 2, …, 8 are 6, 15, 40, 135, 624, 2620, 11160, 48600. The Hasse bound |a| ≤ 2√q held every time.
Figure 8
[known] The leading coefficient here is also elementary: lims→0 s · ζE(s) = N1 / ((1 − q) log q), from the simple pole of the rational function at T = 1. It is the analogue of the 1 / log q on the circle.
What these checks do not do. They do not construct Deninger’s foliated space for Spec Z, and they do not construct a solenoid for these curves. No solenoid constructed, no RH claim. The circle and the elliptic curve are the two cases where the dictionary can be checked directly; they say nothing about whether a space exists for the integers.
A first look at the Camacho–Sad index
[project] The primer above mentions the Camacho–Sad index without computing one. This section computes it on the simplest possible example, so you can see what the number measures. The work was done on 14 November 2026 with exact algebra (sympy), interval arithmetic (mpmath) and numerical contour integration. Every formula below is a [known] result from the literature, re-derived by us on a toy. This is a worked example, not a new theorem, and it makes no claim about the Riemann Hypothesis. The write-up is compute-20261114.md, with the script camacho_sad.py and raw results in camacho-sad-results.json.
The setup, in plain words
[known] Work in two complex variables x and y, and take the vector field X = λ1 x ∂/∂x + λ2 y ∂/∂y, where λ1 and λ2 are nonzero complex numbers. Its only zero is the origin. Its leaves are the curves traced out by the flow, and two of them are the coordinate axes {y = 0} and {x = 0}. A curve like this, which the flow never leaves, is called invariant, or a separatrix.
The Camacho–Sad index of an invariant curve S at a singular point p is a single complex number, CS(S, p), that records how the nearby leaves wind around S. It is defined by a residue (below). You do not need the definition to follow the examples, because in this case the answer is a ratio of the two eigenvalues. What the index buys you is a counting rule. If C is a compact invariant curve, then adding the indices over all singular points on C gives the self-intersection number C·C. That is the Camacho–Sad theorem (1982), and it is the foliation analogue of Poincaré–Hopf from the previous section: a local number at each singular point, with a global total fixed by the topology.
Three results
- [known] Index of an axis. For X = λ1 x ∂x + λ2 y ∂y, the index of {y = 0} at the origin is CS = λ2/λ1, and the index of {x = 0} is λ1/λ2 (Camacho–Sad 1982). [project] We confirmed both with a symbolic residue computation in sympy.
- [known] Blowing up the origin. A blow-up replaces the origin by a whole line E (the set of directions through it) and keeps everything else. E has self-intersection E·E = −1. The foliation now has two singular points on E, one where the direction of the x-axis meets E and one for the y-axis. [project] Our computation gives CS(E) = λ1/(λ2 − λ1) at the first point and λ2/(λ1 − λ2) at the second. Their sum is exactly −1 (sympy), which equals E·E, as the theorem requires. The strict transform of {y = 0} (the axis, lifted to the blow-up) has index λ2/λ1 − 1: its index drops by 1 when we blow up a point on it.
- [project] Numerical support. With mpmath interval boxes for (λ1, λ2), including boxes of width 10−4, the interval for the sum of the two indices on E always contains −1. For complex values (λ1, λ2) = (1 + 0.5i, −0.7 + 2i), contour integration of the residue formula agrees with the closed form to about 30 digits. These checks test our algebra. They are not an independent proof of the Camacho–Sad theorem.
Figure 9
A convention caveat
[known] Sign convention: to be checked against the book. Signs and orientations in the Camacho–Sad index differ between sources, so this matters. The convention used above, which is recalled from memory of Brunella, Birational Geometry of Foliations, Chapter 3 and was not re-read from the book during this stretch, is the following. If S = {f = 0} is invariant and the foliation is given by the 1-form ω = g · df + f · η, then CS(S, p) = −Resp(η/g restricted to S). The only check we ran is internal: this convention gives the classical value λ2/λ1 for {y = 0}, and it makes the two indices on E sum to −1 = E·E. Those are consistent with the theorem, but they do not show that the book’s wording matches ours. Open item: confirm the statement against the text of Brunella and against the original paper before relying on the exact sign.
Reduced versus resonant: a correction
[known] A singular point of a foliation is called reduced (in Seidenberg’s sense) when the linear part of the vector field has eigenvalues λ1, λ2 that are not both zero and whose ratio is not a positive rational number. Seidenberg’s theorem (1968) says that repeated blow-ups turn any singular point of a planar holomorphic foliation into reduced ones. Two different words are easy to confuse here. Resonant means that certain extra terms (Poincaré–Dulac monomials) can survive in the normal form. Non-reduced means the ratio is a positive rational. They are not the same.
[project] Ratio −1/n is already reduced. Take λ1 = n and λ2 = −1. This is resonant, but a negative ratio is reduced, so no blow-up is needed. We computed the indices for n = 1, …, 4: CS({y = 0}) = −1/n and CS({x = 0}) = −n. The first version of our plan for this section described ratio −1/n as the case that needs Seidenberg. That was wrong, and we corrected it. Resonance does not by itself force blow-ups.
[project] A genuinely non-reduced toy: ratio n, with n a positive integer. Take X = x ∂x + (n y + xn) ∂y. For n ≥ 2 this field is in Poincaré–Dulac normal form and is not linearizable, so no change of coordinates makes it a pure linear field. Blowing up the origin (chart y = u x) gives a new singular point whose linear part has ratio n − 1; blowing up again gives n − 2, and so on, down to ratio 1 (the field x ∂x + (x + y) ∂y). After n blow-ups in all, the field has become X = x ∂x + ∂y, which is a regular point, with no singularity left. We verified this chain for n = 1, 2, 3, 4. We followed only the chart that contains the direction of the resonant term, and did not analyse the singularity in the other chart. The index of the invariant curve {x = 0} is 1/n for the whole family, so the resonant term is invisible to it.
[known] This illustrates why Seidenberg’s reduction is needed: a non-reduced point cannot be understood from its linear part alone, and one has to blow up until the model is reduced. It illustrates the mechanism and does not prove Seidenberg’s theorem.
What this section does not do. It does not connect the index to any construction for the integers, and it makes no claim about the Riemann Hypothesis. It checks one textbook formula on one family of examples, with the sign convention still to be confirmed.
A trace formula on the circle
[project] Deninger’s picture asks for a “trace formula” that adds up over closed orbits of a flow and equals a sum over eigenvalues. On the circle toy this formula is not a conjecture: it is Poisson summation, a standard fact. This section writes it down with exact constants and checks it numerically. It is a toy check of a textbook identity. It makes no claim about the Riemann Hypothesis and constructs nothing for the integers. The script is trace_circle.py, with results in trace-circle-results.json (and a correction note for the bump check).
The statement
[known] Let X = ℝ/(log q)ℤ be a circle of length log q, and let the flow translate it at unit speed, so θ = d/dx. Its eigenfunctions are e2πiνx/log q for ν ∈ ℤ, with eigenvalue 2πiν/log q. Its closed orbits are the circle itself, run around k times for k = 1, 2, 3, …, each of length k · log q. Take a test function h on the real line and define its Fourier transform by ĥ(r) = ∫ h(x) e−irx dx. Then
Σν∈ℤ ĥ(2πν / log q) = log q · Σk∈ℤ h(k log q).
[known] This is exactly the Poisson summation formula, Σν ĥ(2πν/T) = T · Σk h(kT), with the period T = log q. The left side is the eigenvalue side: one term per eigenfunction. The right side is the orbit side: one term per number of laps k, and the term at distance k log q is h evaluated there. The k = 0 term, log q · h(0), is the “identity” or volume term. It does not come from a closed orbit. It is the part of the sum that does not see the loop at all, and in Deninger’s dictionary it is the counterpart of the archimedean and main terms in the explicit formula.
What it says, in plain words
Think of a circle as a line with the points x and x + log q glued together. A function on the circle can be described in two ways: by its values, or by how much of each pure wave e2πiνx/log q it contains. Poisson summation says that if you add up a nice function h at evenly spaced points, you get the same total as adding up its Fourier transform at evenly spaced points, provided the spacings are matched (spacing log q on one side, spacing 2π/log q on the other) and the factor log q is included. Spreading h out makes ĥ narrow, and the other way round. That is why the two sides can look so different and still agree: a few wide steps of h correspond to many narrow steps of ĥ.
Gaussian check
[project] We took q = 5, so log q ≈ 1.6094, and h(x) = exp(−x²/(2σ²)), for which ĥ(r) = σ√(2π) · exp(−σ²r²/2) in closed form. We summed both sides to 40 digits with mpmath (the series converge extremely fast). The two sides agree at that precision:
| σ | Eigenvalue side | Orbit side | Difference |
|---|---|---|---|
| 1 | 2.509086351462912758… | 2.509086351462912758… | 0 (< 10−40) |
| 2 | 5.013256549262580502… | 5.013256549262580502… | ≈ 9 × 10−41 |
Eigenvalue side means Σ ĥ(2πν/log 5); orbit side means log 5 · Σ h(k log 5). These are floating-point agreements at 40 digits, not interval-certified enclosures. The nearly exact match for a Gaussian is expected, because the formula is a theorem; the check tests our normalisation (the factor log q and the 2π).
A compactly supported check
[project] For a test function with compact support the orbit side is short. Take h(x) = (1 − x²)² for |x| < 1 and 0 otherwise. Since log 5 ≈ 1.609 is larger than 1, every term with k ≠ 0 vanishes, so the orbit side is just the identity term, log 5 · h(0) = log 5 = 1.609437912434100…. The eigenvalue side is an infinite sum with the closed form ĥ(r) = 16(3 sin r − 3r cos r − r² sin r)/r5 (and ĥ(0) = 16/15). Summing |ν| ≤ 20000 gives 1.609437912434137…, which differs from log 5 by about 3.6 × 10−14. This difference is the truncated tail, which shrinks as the cutoff grows (about 2.7 × 10−7 at |ν| ≤ 100 and 1.8 × 10−10 at |ν| ≤ 1000, from our recomputation of the same sum). So the eigenvalue side converges to the identity term alone, and no closed orbit contributes. An earlier run of this check used an inaccurate numerical integral for ĥ and gave a wrong value (1.65); we replaced it with the closed form, and the correction is recorded in the note.
Figure 10
What this does not do. It checks Poisson summation, which has been known for two centuries, on one circle and two test functions. It does not construct a foliated space for Spec Z, and it does not say what the analogue of the sum over ν would be for the integers. The orbit side here has only one orbit and its repeats; the analogue for ζ(s) would need a sum over primes, and that step is the part Deninger’s programme has not carried out.
What has actually been built
The following items are theorems, or fully written constructions, in the papers cited. They are [known].
- [known] The formal dictionary and the cohomological wish-list in Deninger 1994, 1998, 2001 and 2008. A precise proposal, not a construction of the arithmetic cohomology.
- [known] Function fields, and elliptic curves over finite fields. For a scheme over a finite field one can suspend a discrete Frobenius action to a flow, and closed orbits correspond to closed points with length |o| log p. For an elliptic curve over a finite field, Deninger carries this out with a generalized solenoid, not a manifold. In these examples one can have conformal factor α = 1.
- [known] Leafwise Hodge theory for Riemannian foliations (Álvarez López–Kordyukov, 2001), and a transversal index theorem under strong hypotheses (Álvarez López–Kordyukov; Deninger–Singhof), with Re ρ = α/2.
- [known] Kopei’s product formula for functions meromorphic along the leaves of a three-dimensional foliated flow, the dynamical counterpart of the product formula for a number field.
- [known] A Lichtenbaum analogue for a simple class of isometric flows, assuming regularized determinants exist (Deninger, arXiv:math/0605724), identified with analytic and Reidemeister torsion.
- [known] The circle, as a complete toy. On the circle R / (log q) Z, foliated by points, with R acting by translation, the Ruelle zeta is (1 − q−s)−1, and Lerch’s formula gives the regularized determinant of θ equal to 1 − q−s. The Lichtenbaum-style identities can be checked by hand.
- [known] Leichtnam’s Lefschetz formula for laminated spaces with a p-adic transversal. Under his hypotheses, eigenvalues of θ on a degree-one group have real part 1/2. He states that the existence of the corresponding system for Q is still unknown, except in cases such as elliptic curves over finite fields.
We found no paper in the 1998–2008 literature under a standard name “Kim–Deninger.” The verified constructions around laminated or p-adic-transversal spaces in this programme are Leichtnam’s and the solenoid constructions just cited.
What is open
[known] as open problems stated by the authors themselves.
- The conjectured arithmetic cohomology for schemes over Z has not been constructed (Deninger 2008, §1).
- There is no known foliated dynamical system for Spec Z, or for the ring of integers of a number field, whose leafwise cohomology realises that cohomology. This is the “optimistic guess,” and it remains a guess.
- A transversal index theorem that allows fixed points of the flow (the proposed counterparts of infinite places) and produces the archimedean term of the explicit formula was, at the time of the 2008 survey, ongoing work, not a finished theorem in that paper.
- The identification of a Ruelle zeta with a product of regularized determinants of θ is formal, or is proved under existence assumptions, for a restricted class of isometric flows.
- Leichtnam’s p-adic-transversal formalism does not yet include a system for Q.
[speculative] Any claim that this picture “explains the zeros of ζ,” or that a positivity condition on a leafwise cohomology “is” the Riemann hypothesis, is speculative in the strict sense: the object one would apply the condition to has not been shown to exist. This page does not make that claim.
Sources
A fuller bibliographic list, with keys used in the primer, is in sources.md. The long write-up is primer.md. Primary sources used for this page:
- Christopher Deninger, “Some analogies between number theory and dynamical systems on foliated spaces,” Documenta Mathematica, Extra Volume ICM I (1998), 163–186.
- Christopher Deninger, “Analogies between analysis on foliated spaces and arithmetic geometry,” arXiv:0709.2801 (2007); published 2008 as a chapter in Groups and Analysis: The Legacy of Hermann Weyl, Cambridge University Press, DOI 10.1017/CBO9780511721410.010.
- Christopher Deninger, “Motivic L-functions and regularized determinants,” in Motives, Proc. Sympos. Pure Math. 55, AMS, 1994, 707–743; “Number theory and dynamical systems on foliated spaces,” Jahresbericht der DMV 103 (2001), 79–100; “On the nature of the ‘explicit formulas’ in analytic number theory — a simple example,” in Kanemitsu–Jia (eds.), Number Theoretic Methods, 2002; “A dynamical systems analogue of Lichtenbaum’s conjectures…,” arXiv:math/0605724.
- C. Deninger and W. Singhof, “A note on dynamical trace formulas,” in Dynamical, Spectral, and Arithmetic Zeta Functions, Contemp. Math. 290, AMS, 2001, 41–55.
- J. A. Álvarez López and Y. A. Kordyukov, “Long time behaviour of leafwise heat flow for Riemannian foliations,” Compositio Mathematica 125 (2001), 129–153; “Distributional Betti numbers of transitive foliations of codimension one,” in Foliations: Geometry and Dynamics, 2002, 159–183.
- Eric Leichtnam, arXiv:math/0603576 (Bulletin des Sciences Mathématiques). Fabian Kopei, arXiv:math/0605184.
- A. Candel and L. Conlon, Foliations I, AMS, 2000. C. Camacho and A. Lins Neto, Geometric Theory of Foliations, Birkhäuser, 1985. A. Haefliger, Comment. Math. Helv. 32 (1958), 248–329. M. Brunella, Birational Geometry of Foliations. A. Seidenberg, Amer. J. Math. 90 (1968), 248–269. C. Camacho and P. Sad, Ann. of Math. 115 (1982), 579–595. G. Reeb, Hermann, 1952. H. Hopf, Math. Ann. 96 (1927), 225–250.
- Alain Connes, “Trace formula in noncommutative geometry and the zeros of the Riemann zeta function,” Selecta Math. (N.S.) 5 (1999), 29–106 — related, and not the same construction.
The figures on this page were drawn on 13 November 2026 (Pacific time) for this primer. They are illustrations of published mathematics, not new theorems, and they do not constitute progress on the Riemann Hypothesis.
Reading path · step 10 of 13