Lenses make many points: Pure point spectrum disproves Simon's $L^2$ conjecture

Abstract: For every $d\ge2$ and $0\leσ&lt;1/3$, we construct with LLM-assistance a smooth bounded nonnegative lens-type potential $V$ with arbitrarily small $$|V|<em>\infty+\int</em>{\mathbb{R}^{d}|x|^{{1-d+σ}|V(x)|²}} dx$$, such that $-Δ+V$ has only pure point spectrum in $(0,\infty)$. Thus, we show that Barry Simon's $L<sup>2$ conjecture, Problem 8 in "Schrödinger operators in the twenty-first century", is false.

• The paper demonstrates that for every $d ransition{title}:ge 2$, there exist smooth, nonnegative potentials with arbitrarily small $\{V\}$ ransition{math}$\_\infty$, $\]I ransition{math}\_d ranisiton{Moving}$ ransition{subtitle}_0(.)}, and $\]I ransition{math}\_d ransition{Power}0 υ$ with pure point spectrum in Schrödinger operators.
• The construction uses annular lenses and barriers to create barriers supporting compactly supported approximate solutions on a prescribed bounded interval $({2 transit}3 ransition{empty}Pets”) and scales exponential decay.
• A global spectral statement is derived from local block estimates using spectral-averaging criterion, ensuring the existence of a bounded operator $F\_\lambda$ with pure point spectrum.

Problem and principal result

The paper addresses Barry Simon’s multidimensional $L^2$ conjecture for Schrödinger operators

$H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$

which asserts, in particular, that the weighted condition

$$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$

forces an absolutely continuous spectral component of infinite multiplicity at almost every positive energy. The conjecture was motivated by the one-dimensional Deift–Killip theorem and by classical scattering results under pointwise decay assumptions.

The main theorem gives a substantially stronger counterexample. For every $d\ge 2$, every $0\le \sigma<1/3$, and every $\delta>0$, the authors construct a smooth, bounded, nonnegative potential $V$ satisfying

$$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$

such that

$$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$

Thus the operator has no absolutely continuous or singular continuous spectrum on the positive half-line. Its spectrum there is pure point, with eigenfunctions forming a complete orthonormal basis; in dimensions two and three, the positive eigenvalues have finite multiplicity and are dense in $[0,\infty)$ (2610.31637).

The contradiction with Simon’s conjecture is therefore not caused merely by the appearance of singular continuous spectrum. The construction eliminates the entire continuous spectral subspace while preserving the essential spectrum of the free Laplacian and while making the potential arbitrarily small in both the supremum norm and a stronger weighted $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$0 scale.

The geometric mechanism: annular lenses and barriers

The two-dimensional construction is a sparse sum of smooth potentials supported on widely separated annuli. At each scale, an annular block contains two components: an angular lens and a radial reflecting barrier. The lens creates narrow angular channels, while the barrier is arranged to reflect waves concentrated in those channels. The resulting block supports compactly supported approximate solutions with arbitrarily small residuals for most energies in a prescribed bounded interval.

The scale separation is essential. If a source is supported in a disk of radius $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$1, the next annular block is placed at radius $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$2. The parameters are chosen using

$H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$3

together with

$H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$4

Here $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$5 controls the angular lens strength, $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$6 is the angular frequency scale, and $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$7 is the barrier thickness. The restriction $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$8 appears directly in the positivity of $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$9 and hence in the decay of the weighted cost of each block.

The lens is generated by an angular potential of the form

$$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$0

where the parameter $$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$1 is slowly increased from $$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$2 to $$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$3 along the annulus. At the terminal region, the ground state of the angular operator is concentrated in an interval of angular width approximately $$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$4, with $$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$5. The barrier is then multiplied by an angular cutoff supported in this narrow region.

Figure 1: Three illustrative annular blocks of the potential, showing the sparse radial placement of the lens–barrier structures.

The quantitative cost estimate is one of the central points of the construction. The lens and barrier contributions satisfy

$$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$6

and, with the selected exponents,

$$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$7

Consequently, each block has vanishing weighted cost as the scale tends to infinity, uniformly in the energy ceiling used to construct the barrier. This uniformity permits a single sparse potential to control every compact positive-energy interval.

Angular confinement and adiabatic transport

The lens analysis reduces the angular problem to the family

$$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$8

where $$\mathcal I_{d,0}(V) =\int_{\mathbb R^d}|x|^{1-d}|V(x)|^2\,dx<\infty$$9 is the Floquet parameter associated with the angular Fourier mode. The authors prove a uniform spectral gap

$d\ge 2$0

for $d\ge 2$1. The ground eigenvalue satisfies

$d\ge 2$2

and for $d\ge 2$3 the corresponding normalized ground state is exponentially concentrated near the angular channel: $d\ge 2$4

This growing gap allows the authors to transport the angular ground state through the radial lens using an adiabatic expansion. The radial variation introduces a small parameter

$d\ge 2$5

and the profile is corrected to order $d\ge 2$6. The resulting approximate angular-radial waves have residual bounded by

$d\ge 2$7

uniformly over all angular channels $d\ge 2$8 and energies $d\ge 2$9. Increasing $0\le \sigma<1/3$0 makes the residual smaller than any prescribed negative power of $0\le \sigma<1/3$1.

The use of the entire Fourier ladder is important. The estimates are not restricted to a single angular mode or to a truncated angular approximation: the profile construction controls all Fourier modes in the auxiliary angular variable. This is necessary for the subsequent matching argument, where arbitrary $0\le \sigma<1/3$2 sources must be treated by a common approximate inverse.

Reflecting barriers and compactly supported approximate inverses

The radial barriers are assembled from finitely many smooth periodic potentials whose spectral gaps cover a prescribed interval $0\le \sigma<1/3$3. Their heights remain bounded by $0\le \sigma<1/3$4, independently of the size of $0\le \sigma<1/3$5 once the scale is sufficiently large. For every energy in the relevant interval, one periodic component has an exponential dichotomy, producing a decaying solution. A long concatenation of these components yields a compactly supported quasimode whose residual is exponentially small in the barrier thickness: $0\le \sigma<1/3$6

The lens waves and the decaying barrier solutions are matched at the outer edge of the lens. At the inner edge, they are matched to free Hankel waves. The authors establish estimates uniform in the growing angular range $0\le \sigma<1/3$7 using large-argument Hankel asymptotics. The reflected waves have the form

$0\le \sigma<1/3$8

and carry $0\le \sigma<1/3$9 mass comparable to $\delta>0$0 in the annular region: $\delta>0$1

A finite-volume Dirichlet operator is then used to exclude a small set of energies near its eigenvalues. If $\delta>0$2 avoids this exceptional set, the matching denominators are bounded below, and every source supported in $\delta>0$3 admits a compactly supported approximate solution $\delta>0$4 satisfying

$\delta>0$5

and

$\delta>0$6

The excluded set is common to all sources $\delta>0$7 and has measure at most $\delta>0$8 after choosing the construction order appropriately. This commonality is crucial: a source-dependent exceptional set would not suffice for the global resolvent argument.

From local approximate inverses to pure point spectrum

The local block estimates are converted into a global spectral statement by a spectral-averaging criterion due to Combes and Hislop, in the formulation used by Fischer, Leschke, and Müller. The abstract assembly proposition considers potentials $\delta>0$9 supported on annuli with radii satisfying

$V$0

If compactly supported approximate inverses have residual size $V$1 outside exceptional sets of summable measure, the finite-volume resolvent comparison yields a global weighted right inverse.

The key exponent conditions are

$V$2

The first condition makes the finite-volume spectral neighborhoods summable. The second makes the errors between successive scales summable. The third ensures that the weighted assembly operator is bounded.

More precisely, for almost every energy $V$3 in a bounded interval $V$4, the construction produces a bounded operator $V$5 such that

$V$6

It follows that

$V$7

The multiplication operator $V$8 has dense range, and the sandwiched resolvent is compact by Rellich compactness. Spectral averaging applied to

$V$9

then gives pure point spectrum on $$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$0 for almost every $$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$1. A countable exhaustion of $$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$2 produces a single coupling for which the conclusion holds at every positive energy.

The perturbation $$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$3 is subsequently included in the potential. Because $$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$4 can be chosen arbitrarily small, it does not affect the smallness of the weighted norms. Nonnegativity implies that the kernel is trivial: $$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$5 forces $$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$6.

Extension to three and higher dimensions

The three-dimensional construction uses the planar potential as a transverse background. The authors add two disk-shaped barriers in the remaining coordinate, producing a long finite cylinder. The axial barriers are again built from periodic pieces with spectral gaps covering a prescribed energy interval. A product decomposition separates the transverse planar operator from the axial Dirichlet operator.

The main technical issue is uniformity over all axial eigenvalues. For a separated mode with axial eigenvalue $$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$7, the transverse energy is

$$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$8

which may be arbitrarily negative. The planar inverse is therefore supplemented by a low-energy inverse obtained from the angular lens alone. An Agmon estimate gives exponential suppression near the outer planar boundary for energies $$\|V\|_\infty+\mathcal I_{d,0}(V)+\mathcal I_{d,\sigma}(V)<\delta$$9. This avoids any assumption that the separated energy remains in a fixed positive interval.

The cylindrical block produces compactly supported approximate inverses with residual

$$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$0

outside an exceptional set of measure $$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$1. The selected assembly parameters are

$$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$2

for which

$$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$3

The weighted cost of one pair of caps satisfies

$$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$4

which is summable in the chosen sparse sequence for $$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$5.

For arbitrary dimensions, the authors take tensor sums of two- and three-dimensional operators. The tensor-product eigenfunctions form a complete orthonormal basis, and the spectrum consists of sums of factor eigenvalues. The resulting operator still has pure point spectrum, no kernel, and spectrum equal to $$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$6. A final dilation

$$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$7

preserves the spectral type while giving

$$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$8

so all required smallness conditions can be imposed simultaneously.

Limitations and open questions

The threshold $$\sigma(H_V)=\sigma_{\mathrm{ess}}(H_V)=[0,\infty), \qquad P_c(H_V)=0.$$9 is intrinsic to the parameter balance proved in this construction, but the paper does not establish that $[0,\infty)$0 is optimal. It remains open whether analogous counterexamples exist for $[0,\infty)$1, or whether a stronger weighted condition eventually forces a continuous spectral component.

The theorem is also existential and highly structured. The potential is nonnegative, smooth, bounded, and sparse, but its construction depends on increasingly separated annular or cylindrical blocks, energy ceilings, exceptional-set estimates, and a spectral-averaging parameter. The argument therefore does not characterize typical potentials satisfying the weighted $[0,\infty)$2 condition. It shows only that that condition, even with the additional factor $[0,\infty)$3, cannot by itself imply absolutely continuous spectrum.

The pure point conclusion is obtained after adding a small decaying perturbation $[0,\infty)$4. Although the perturbation can be made arbitrarily small in the relevant norms, the proof establishes the spectral statement for almost every coupling $[0,\infty)$5, not for every $[0,\infty)$6. The dependence of the exceptional coupling set on the detailed block construction is not analyzed further.

Finally, the argument explicitly records assistance from ChatGPT in the discovery and development of the construction, while assigning mathematical responsibility to the authors. The substantive proof is nevertheless rewritten in terms of independently stated spectral, elliptic, adiabatic, and matching estimates; the paper does not provide a general methodology for verifying machine-generated arguments in this area.

Conclusion

The paper disproves Simon’s multidimensional $[0,\infty)$7 conjecture by constructing, in every dimension $[0,\infty)$8, smooth nonnegative potentials with arbitrarily small $[0,\infty)$9, $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$00, and $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$01 for every $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$02, while the corresponding Schrödinger operator has spectrum $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$03 and no continuous spectral component. The proof combines angular confinement, adiabatic lens transport, periodic-gap barriers, finite-volume matching, resolvent assembly, and spectral averaging. Its principal conclusion is that even a strengthened weighted $H_V=-\Delta+V\quad\text{on }L^2(\mathbb R^d),$04 condition does not guarantee the persistence of absolutely continuous spectrum in dimensions two and higher (2610.31637).