McKean-Type Spectral Gap Estimate

• McKean-type spectral gap estimates are methods that establish positive lower bounds on the first nonzero eigenvalue using global geometric invariants such as negative curvature and averaged Ricci curvature.
• They extend classical Laplacian bounds by incorporating averaged and integral curvature conditions, applicable in contexts from Riemannian manifolds to Finsler spaces and random covers.
• The approach has wide applications, providing control over eigenvalue distributions in quantum graphs, discrete structures, and operator-theoretic settings, ensuring a positive spectral floor.

A McKean-type spectral gap estimate is a lower bound on the bottom of the positive spectrum—most often the first nonzero eigenvalue, the bottom of the $L^2$-spectrum, or a fundamental tone—derived from global geometric or structural input. In the strict historical sense, the model is McKean’s lower bound for negatively curved manifolds; in contemporary usage on arXiv, the phrase also covers analogues in which sectional curvature is replaced by averaged Ricci data, universal-cover thresholds, Finsler flag curvature, projective contraction, Wasserstein contraction, combinatorial expansion, or higher-dimensional simplicial structure (Bonnefont et al., 2021, Hide et al., 15 Feb 2025, Kajántó, 19 Aug 2025, Han et al., 2019, Hoffman et al., 2012).

1. Classical geometric template and manifold extensions

In the classical Riemannian setting, the McKean paradigm concerns complete negatively curved manifolds and lower bounds for the bottom of the Laplace spectrum. One formulation recalled in recent work is the standard estimate for a complete simply connected $n$-manifold with sectional curvature bounded above by a negative constant,

$$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$

while for negatively curved surfaces with curvature bounded by

$-b^2\le K_{\widetilde g}(z)\le -a^2<0,$

McKean’s theorem yields

$\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$

These formulas identify the basic structural pattern: sufficiently strong negative curvature forces a strictly positive spectral floor (Bonnefont et al., 2021, Hide et al., 15 Feb 2025).

Recent manifold results preserve that pattern but weaken the hypotheses. On complete Riemannian manifolds, lower bounds can be expressed through the smallest eigenvalue $\rho(x)$ of the Ricci tensor, its positive part averaged on balls,

$\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$

and a form-boundedness condition on $\rho_-$. Under these assumptions one obtains

$$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$

in the infinite-volume case, and

$$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$

in the finite-volume case, together with explicit lower bounds involving covering multiplicities and local Neumann eigenvalues. This is McKean-type in the spectral sense—strict positivity of the bottom of the spectrum—but the mechanism is averaged Ricci positivity rather than sectional-curvature negativity (Bonnefont et al., 2021).

A different compact-manifold extension replaces pointwise Ricci lower bounds by a small integral deficit below $n$0. If

$n$1

is sufficiently small, with $n$2, then for every $n$3,

$n$4

where $n$5 is the first nontrivial eigenvalue of the one-dimensional Bakry–Qian/Kröger model on $n$6. In the negative-curvature regime $n$7, this implies the explicit hyperbolic-type lower bound

$n$8

This is not a literal noncompact McKean theorem, but it is the compact diameter-controlled analogue in that framework (Olivé et al., 31 Oct 2025).

2. Random covers and the universal-cover threshold

A probabilistic version of the McKean philosophy appears in the Laplace spectrum of large random finite covers of a fixed negatively curved compact surface. Let $n$9 be a compact connected surface of strictly negative curvature, $$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$0 its universal cover, $$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$1 the bottom of the $$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$2-spectrum on $$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$3, and

$$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$4

which equals the volume entropy of $$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$5 and the topological entropy of the geodesic flow on $$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$6. For a random $$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$7-sheeted cover $$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$8, the main theorem defines

$$K_M\le -k^2<0 \quad\Longrightarrow\quad \lambda_0(M)\ge \frac{(n-1)^2k^2}{4},$$9

and proves that for every $-b^2\le K_{\widetilde g}(z)\le -a^2<0,$0, with high probability as $-b^2\le K_{\widetilde g}(z)\le -a^2<0,$1,

$-b^2\le K_{\widetilde g}(z)\le -a^2<0,$2

with multiplicities agreeing. Equivalently, the first new eigenvalue $-b^2\le K_{\widetilde g}(z)\le -a^2<0,$3 satisfies

$-b^2\le K_{\widetilde g}(z)\le -a^2<0,$4

The estimate is therefore relative to the base surface: inherited eigenvalues are not excluded, but new spectrum is pushed above a universal-cover-based threshold (Hide et al., 15 Feb 2025).

The McKean connection is explicit. Since negative curvature implies $-b^2\le K_{\widetilde g}(z)\le -a^2<0,$5 and Bishop comparison gives $-b^2\le K_{\widetilde g}(z)\le -a^2<0,$6, the threshold

$-b^2\le K_{\widetilde g}(z)\le -a^2<0,$7

is positive whenever

$-b^2\le K_{\widetilde g}(z)\le -a^2<0,$8

In the hyperbolic case,

$-b^2\le K_{\widetilde g}(z)\le -a^2<0,$9

so the theorem yields

$\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$0

The same paper formulates the conjecturally optimal statement

$\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$1

with high probability, and proves near-optimal existence along deterministic sequences of covers by strong convergence methods. A plausible interpretation is that $\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$2 plays the role of a Ramanujan threshold for random covers in variable negative curvature (Hide et al., 15 Feb 2025).

3. Finsler and irreversible generalizations

The Finsler version of a McKean-type estimate replaces sectional curvature by flag curvature and introduces irreversibility as an essential extra parameter. For a forward complete, non-compact Finsler metric measure manifold $\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$3 of dimension $\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$4 and finite reversibility $\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$5, let $\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$6. If along $\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$7,

$\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$8

then for every $\lambda_0(\widetilde X)\ge \frac{a^2}{4}.$9 and every $\rho(x)$0,

$\rho(x)$1

where

$\rho(x)$2

For $\rho(x)$3, this is the direct spectral-gap analogue: $\rho(x)$4 The estimate is variational rather than operator-theoretic in presentation, but it controls the bottom of the Dirichlet spectrum in the natural $\rho(x)$5-energy sense (Kajántó, 19 Aug 2025).

The distinctive feature is sharpness in a genuinely irreversible family. The paper constructs

$\rho(x)$6

a family of projectively flat Randers metric measure manifolds on the Euclidean ball with

$\rho(x)$7

for which

$\rho(x)$8

and $\rho(x)$9 is the greatest constant with this property. There are no genuine extremizers in $\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$0; sharpness is attained by truncated radial minimizing sequences built from the formal profile

$\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$1

This distinguishes the result from reversible Hardy-type theory, where sharpness questions are much more classical (Kajántó, 19 Aug 2025).

4. Comparison, contraction, and operator-theoretic analogues

Outside negative-curvature geometry, the phrase “McKean-type” is often used analogically for comparison principles that force a gap from a one-dimensional model, a contraction coefficient, or an explicit tail criterion. For Euclidean Schrödinger operators on bounded smooth strictly convex domains,

$\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$2

with Dirichlet boundary conditions, one sharp form is

$\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$3

where $\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$4 is a one-dimensional modulus of convexity of $\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$5 on $\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$6. In particular, when $\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$7 is convex,

$\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$8

This is not McKean’s curvature theorem, but it is comparison-theoretic in exactly the same structural sense: geometric convexity is replaced by a one-dimensional model operator controlling the first positive spectral separation (He, 2014).

A diffusion analogue on $\delta(R):=\inf_{p\in M}\fint_{B(p,R)} \rho_+(x)\,dx>0,$9 is provided by spherically symmetric log-concave measures

$\rho_-$0

for which the associated symmetric generator satisfies

$\rho_-$1

The lower bound comes from a radial-angular decomposition and a one-dimensional estimate for the radial part. This gives a dimension-sharp Euclidean analogue of Lichnerowicz/McKean-type lower bounds, although the setting is probabilistic rather than manifold-geometric (Bonnefont et al., 2014).

In the Perron–Frobenius setting, a strictly positive matrix $\rho_-$2 has Perron eigenvalue $\rho_-$3 and spectral ratio

$\rho_-$4

Using a complex extension of Hilbert’s projective metric, one obtains

$\rho_-$5

where $\rho_-$6 is the Birkhoff contraction coefficient, equivalently

$\rho_-$7

Hence

$\rho_-$8

This is McKean-type only by analogy: positivity and projective contraction replace curvature, but the outcome is still a computable gap lower bound from a global geometric quantity (Han et al., 2019).

For Markov operators, the abstract criterion becomes even more operator-theoretic. If $\rho_-$9 is an ergodic Markov operator on $$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$0 and

$$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$1

then $$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$2 has a spectral gap if and only if

$$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$3

Equivalent formulations use odd powers of $$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$4 or the tail norm

$$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$5

This is not an explicit lower bound in the form $$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$6, but it gives an exact necessary-and-sufficient spectral-gap criterion and implies that defective Poincaré or defective log-Sobolev-type inequalities are equivalent to their tight versions in the symmetric irreducible Dirichlet-form setting (Wang, 2013).

A probabilistic contraction version appears in simple slice sampling. For rotationally invariant log-concave targets,

$$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$7

with $$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$8 strictly increasing and convex, the one-step kernel satisfies the Wasserstein contraction

$$s(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$9

hence

$$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$0

More generally, if the level-set volume profile $$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$1 belongs to $$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$2, then

$$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$3

This suggests a McKean-type principle in coarse Ricci form: one-step $$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$4-contraction yields an explicit $$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$5 spectral gap (Natarovskii et al., 2019).

A related local-to-global principle governs the weighted adjacent-transposition chain on $$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$6. If

$$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$7

then every positive eigenvalue of the transition matrix $$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$8 is at least

$$\lambda_1(\Delta)\ge s(\Delta+\rho)\ge (1-a)s(\Delta+\rho_+)$$9

In the regular case $n$00, so

$n$01

and this is sharp because equality holds for the uniform chain. The mechanism is projection-theoretic rather than geometric, but it remains a local-obstruction-to-global-gap theorem (Greaves et al., 27 Mar 2026).

5. Graphs, random complexes, and discrete expansion

In discrete probability, one important analogue is stronger than a mere lower bound on $n$02: the entire nonzero spectrum may concentrate. For the normalized graph Laplacian

$n$03

of an Erdős–Rényi graph, if

$n$04

then on the giant component $n$05,

$n$06

with probability at least

$n$07

Equivalently,

$n$08

so in particular

$n$09

Below the sharp $n$10-threshold, this concentration fails: if

$n$11

then

$n$12

with high probability. The paper explicitly states that this is stronger than a classical McKean estimate: not only is the positive spectrum bounded away from $n$13, it is asymptotically pinned near the midpoint $n$14 of the normalized-Laplacian spectrum (Hoffman et al., 2012).

For higher-dimensional simplicial Laplacians, the smallest eigenvalue

$n$15

plays the role of the spectral gap. Two comparison estimates are central. For a subcomplex $n$16,

$n$17

where $n$18 counts missing $n$19-faces adjacent to a $n$20-face. For a general simplicial complex,

$n$21

where the defect parameters $n$22 quantify failures of flagness. In the clique-complex case these defects vanish and one recovers

$n$23

Because $n$24, positivity of these lower bounds implies vanishing of reduced cohomology. This is McKean-type in a higher-dimensional combinatorial sense: the bottom of the Laplacian is controlled by expansion of the $n$25-skeleton plus explicit defect terms (Shukla et al., 2018).

A different discrete phenomenon is complementarity. For the transition matrix of simple random walk on a graph $n$26, the random-walk spectral gap is

$n$27

equivalently the first positive eigenvalue of the normalized Laplacian. The Nordhaus–Gaddum theorem proved in this setting is

$n$28

If all degrees satisfy

$n$29

then

$n$30

Thus one graph and its complement cannot both have arbitrarily small gap under dense-degree hypotheses. This is not curvature-based, but it is unmistakably McKean-like in the sense of a universal lower bound on a first nontrivial eigenvalue under coarse structural control (Kim et al., 2024).

6. Quantum graphs, boundary cases, and conceptual distinctions

Quantum graphs reveal a limitation of the McKean analogy: no lower bound depending only on diameter is possible for the Kirchhoff Laplacian on compact metric graphs. There exist flower-dumbbell graphs with fixed diameter $n$31 and

$n$32

and pumpkin chains with the same fixed diameter and

$n$33

What survives are mixed-parameter estimates. For every compact connected metric graph of total length $n$34,

$n$35

with equality for the path graph. With diameter $n$36 and number of edges $n$37,

$n$38

With diameter $n$39 and total length $n$40,

$n$41

hence in particular

$n$42

The paper’s principal negative result is therefore that diameter alone is insufficient as a McKean-type control parameter in the quantum-graph category (Kennedy et al., 2015).

A more positive quantum-graph analogue comes from transference to discrete normalized Laplacians. If $n$43 is an $n$44-fold cover of a metric graph $n$45, $n$46 the associated vicinity graph, and

$n$47

then

$n$48

where $n$49 are the eigenvalues of the weighted normalized Laplacian of $n$50. In particular,

$n$51

For planar bridgeless graphs with a cycle double cover and dual graph $n$52, this becomes

$n$53

Here the McKean-like mechanism is discrete expansion of a vicinity or dual graph combined with local lower bounds on the spectral gaps of the covering pieces (Mugnolo et al., 2019).

A recurring misconception is that every paper with “McKean” in its conceptual background concerns a lower bound on the first positive eigenvalue. That is false. In graph theory, the McKean–Singer formula

$n$54

and more generally

$n$55

describe supersymmetric cancellation of nonzero spectrum between even and odd form sectors. They do not provide a lower bound of the form $n$56. The distinction matters: McKean–Singer is an index-theoretic supertrace identity, whereas a McKean-type spectral gap estimate is a coercive lower bound on the bottom of the positive spectrum (Knill, 2013).

Taken together, these developments suggest that “McKean-type spectral gap estimate” is now best understood as a structural category rather than a single theorem. In its literal form it refers to negative-curvature lower bounds for spectral bottoms; in modern extensions it includes universal-cover thresholds, integral-curvature comparisons, irreversible Finsler inequalities, contraction-based operator bounds, and combinatorial or high-dimensional expansion estimates. The unifying feature is always the same: a global geometric or structural constraint excludes small positive spectrum.