Pith. sign in

REVIEW 4 major objections 5 minor 13 references

Multiscale light-matter dynamics in quantum materials: from electrons to topological superlattices

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A single multiscale simulation code sustains 1.87 EFLOP/s and cuts quantum-dynamics time-to-solution 152x versus the prior record.

desk verdict The integration is real but the headline numbers are not: the 3,780x speedup is a metric artifact and the 100.2% of peak is internally inconsistent. read the letter →

arxiv 2509.00966 v1 pith:4HSFG7SQ submitted 2025-08-31 physics.comp-ph

Taufeq Mohammed Razakh , Thomas Linker , Ye Luo , Nariman Piroozan , John Pennycook , Nalini Kumar , Albert Musaelian , Anders Johansson
show 7 more authors
Boris Kozinsky Rajiv K. Kalia Priya Vashishta Fuyuki Shimojo Shinnosuke Hattori Ken-ichi Nomura Aiichiro Nakano
This is my paper · ORCID
classification physics.comp-ph
keywords light-matterdynamicsquantummoleculardivide-and-conqueralgorithmsexascalecomputingneural-networkpotentialsmixed-precisionarithmetictopologicalmaterialsferroelectricsuperlattice
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that the full multiscale chain of light-matter dynamics—Maxwell's equations for light, Schrödinger equations for electrons, Newton's equations for atoms, and neural-network surrogate dynamics—can be simulated end to end on an exascale machine by decomposing the problem physically as well as spatially and mapping each piece to the hardware that suits it. On a 10,000-node exascale system it reports sustaining 1.873 EFLOP/s (100.2% of FP64 peak) for a 15,360,000-electron quantum-dynamics run, along with 152x and 3,780x reductions in time-to-solution for quantum dynamics and excited-state neural-network molecular dynamics versus prior state-of-the-art results. These numbers, if correct, make this the first end-to-end exascale multiscale light-matter simulation and enable the first first-principles study of light-induced switching in a ferroelectric skyrmion superlattice. The broader claim is that hardware heterogeneity and low-precision arithmetic can be used as resources rather than obstacles for scientific computing.

What carries the argument

The load-bearing mechanism is the divide-conquer-recombine split into physical subproblems with small dynamic ranges: local field dynamics—light-driven Schrödinger and Maxwell propagation on regular grids—runs on GPU, while quantum excitation molecular dynamics—nonadiabatic electron-atom coupling and surface hopping—runs on CPU. Shadow dynamics and metamodel-space algebra let the two exchange only tiny occupation-number deltas, amortized over roughly a hundred quantum-dynamics steps, so large wave functions stay GPU-resident. Nonlocal corrections are 'GEMMified' into dense matrix multiplications that tolerate BF16 precision; excited-state neural-network molecular dynamics blends a ground-sta

What would settle it

Instrument the 15,360,000-electron run with an independent FLOP counter that separates FP64, FP32, and BF16 operations and divide by wall-clock time; if the result is well below 1.87 EFLOP/s, the peak-fraction claim fails. Also rerun the previous state-of-the-art benchmark—71,040 electrons, comparable accuracy, same time-to-solution definition—with this code; if the time-to-solution is not near the claimed 1.11e-7 seconds per electron per step, the 152x comparison is configuration-dependent.

Watch

Extended reading notes

Core claim

The central claim is that a divide-conquer-recombine algorithmic paradigm, combined with a metamodel-space algebra that minimizes inter-unit communication, turns the exascale heterogeneity challenge into a speed advantage. The paper decomposes the coupled light-electron-atom problem into local field dynamics run on GPU tiles and quantum excitation molecular dynamics run on CPUs, exchanging only small occupation-number changes instead of large wave-function arrays. Nonlocal corrections are reframed as dense matrix multiplications and run in FP32/BF16, with FP64 kept for the chemically sensitive part. On 10,000 nodes (60,000 GPUs), this yields 1.873 EFLOP/s for a 15,360,000-electron PbTiO3 sys

Load-bearing premise

The headline speedups rest on the assumption that time-to-solution normalized by electron count or atom-weight is a fair way to compare runs on different machines with very different models; if the reference runs are not equivalently configured, the speedup factors lose their meaning.

Editorial extensions

If this is right

  • Quantum light-matter dynamics of multimillion-electron systems becomes practical on exascale machines, not just ground-state electronic structure.
  • The divide-conquer-recombine/metamodel-space algebra pattern provides a template for mapping other multiscale and multiphysics problems onto heterogeneous, low-precision hardware.
  • Excited-state molecular dynamics of trillion-atom systems with first-principles-accuracy neural networks becomes feasible, enabling long-time simulations that were previously out of reach.
  • Simulation-guided study of light-induced topological switching can proceed at device-relevant scales, potentially informing experiments on ferroelectric topotronic devices.
  • Mixed-precision arithmetic is sufficient for the local part of quantum propagation, while full FP64 is reserved for chemically sensitive parts—a precision hierarchy for future exascale applications.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The divide-conquer-recombine pattern suggests a recipe for other multiphysics codes: decompose by dynamic range, pair each subproblem with its best precision, and exchange only low-dimensional summaries at coarse time steps; this could be tested on couplings such as fluid-plasma or chemistry-atmosphere systems.
  • The shadow-dynamics exchange of occupation numbers rather than wave functions implies that multi-fidelity AI/physics handshaking can work when the AI module receives only a scalar summary of the quantum state, which may generalize to other neural-network-accelerated simulations.
  • The reported system sizes open a concrete comparison target: experimental pump-probe measurements of ferroelectric skyrmion switching should now be simulable at comparable length and time scales, so threshold fluences and switching times can be checked directly.
  • Because the code uses OpenMP target offload and standard BLAS calls, porting it to a different vendor's GPUs is a direct test of whether the heterogeneity exploitation is portable or specific to the reported machine.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents MLMD, a multiscale light-matter dynamics software integrating the DC-MESH (divide-and-conquer Maxwell–Ehrenfest–surface hopping) and XS-NNQMD (excited-state neural-network quantum molecular dynamics) modules. On the Aurora supercomputer, the authors report 1.873 EFLOP/s for a 15.36-million-electron PbTiO3 system, a 152-fold time-to-solution (T2S) improvement over prior Maxwell–Ehrenfest simulations, a 3,780-fold T2S improvement for a 1.23-trillion-atom XS-NNQMD simulation, and near-perfect weak scaling for both modules. The paper claims this is the first end-to-end exascale multiscale light-matter dynamics simulation and demonstrates light-induced switching of topological superlattices.

Significance. If the headline numbers were sound, this would be a significant Gordon Bell–level achievement: integrating Ehrenfest and surface-hopping dynamics with neural-network potentials at exascale, exploiting mixed precision and heterogeneous hardware, and reporting near-perfect weak scaling. The algorithmic contributions—DCR spatial/physical decomposition, shadow dynamics, GEMMification of nonlocal corrections, and parameterized mixed precision—are interesting and the per-kernel optimization data are valuable. However, the central performance claims are undermined by two load-bearing problems: the XS-NNQMD 3,780× speedup is an artifact of a weight-normalized T2S metric, and the 1.873 EFLOP/s figure is reported as 100.2% of FP64 peak while the stated machine FP64 peak is approximately 2 EFLOP/s. The scientific claim about topological superlattice switching is also not substantiated with results. As written, the paper's headline claims do not survive scrutiny.

major comments (4)
  1. [Sec. IV, Sec. VII.C.2, Table II] The 3,780× XS-NNQMD speedup is manufactured by the T2S definition. The paper defines T2S as wall-clock time per MD step divided by (number of atoms × number of neural-network weights). With the authors' own numbers, SOTA [11] gives 3142.66 s/(1.007e12 atoms × 440 weights)=7.09e-12 s/(atom·weight), while this work gives 1590.31 s/(1.2288e12 atoms × 690,000 weights)=1.88e-15 s/(atom·weight), a ratio of ~3,780. If one instead compares the actual wall-clock time per atom per step—the only quantity that reflects time-to-solution—the SOTA is 3.12e-9 s/(atom·step) and this work is 1.29e-9 s/(atom·step), a speedup of only ~2.4×. Dividing by the number of weights rewards models with more parameters and has no direct connection to accuracy or time-to-solution. Since the abstract and Gordon Bell justification rest on the 3,780× number, this is a load-bearing flaw.
  2. [Sec. VII.B, Sec. VI.B, Table I] The peak-performance claim is internally inconsistent. The paper states in Sec. VI.B that the entire Aurora machine has a theoretical FP64 peak of approximately 2 EFLOP/s, yet Sec. VII.B reports 1.873 EFLOP/s on 10,000 nodes as "100.2% of FP64 peak." 1.873/2 = 93.65%, not 100.2%. If the denominator is instead the FP64 peak of the allocated 10,000 nodes, that should be stated explicitly, along with how that peak is computed. Additionally, Table IV lists a single-tile FP64 peak of 23 TFLOP/s, while Sec. VI.B says a blade has 187 TFLOP/s peak; with 6 GPUs × 2 tiles per blade, these numbers do not reconcile. The FLOP accounting (Intel SDE, unitrace, scaling from a single tile to 10,000 nodes) needs clarification before the 1.87 EFLOP/s claim can be assessed.
  3. [Sec. VII.C.1, Table I] The 152× DC-MESH T2S improvement is also not an apples-to-apples comparison. The paper compares 1.11e-7 s/electron for a 15,360,000-electron PbTiO3 MESH simulation on 10,000 Aurora nodes with 1.69e-5 s/electron for a 71,040-electron silica Maxwell–Ehrenfest subset simulation on 27,648 Fugaku nodes. These runs differ in problem size by more than two orders of magnitude, in physical system, and in the physics included (MESH superset vs ME subset). The large speedup is at least partly a consequence of scale and machine generation. A convincing claim would require either same-machine, same-size controlled comparisons or a clear argument that the per-electron T2S is insensitive to these differences.
  4. [Sec. V.B.7, Sec. VI.C, Sec. VII.B] The mixed-precision accuracy claim is asserted rather than demonstrated. The paper says BF16 with FP32 accumulation is sufficient for the nonlocal correction and refers to [34], but no error analysis or validation is provided for the production 15.36-million-electron run that achieves 1.873 EFLOP/s. Since the FLOP/s number and the scientific validity of the dynamics both depend on this precision choice, the paper should include a direct convergence check (e.g., comparing BF16/FP32 results against FP64 for observables such as electronic occupations or excited-state populations).
minor comments (5)
  1. [Abstract and Performance Attributes] Several superscripts are garbled, e.g., "1.88×10!"# [sec/(atom•weight•step)]". Please fix all exponent and unit formatting.
  2. [Sec. VI.B] The hardware description is ambiguous: "PVC tiles per blade have a peak floating-point performance of 187 TFLOP/s" is unclear whether 187 TFLOP/s is per tile, per GPU, or per blade. The later statement that machine peak is ~2 EFLOP/s implies per blade, but Table IV uses 23 TFLOP/s per tile. Please clarify.
  3. [Sec. VIII and Fig. 3] The paper claims to have "enabled the first study of light-induced switching of topological superlattices," but no switching dynamics, order-parameter time series, or comparison with experiment is presented. As a performance paper this may be outside scope, but the scientific claim should be toned down or supported.
  4. [Author affiliations] Typo: "Insitute" should be "Institute" in the Stanford PULSE affiliation.
  5. [Sec. IV and Table II] The SOTA XS-NNQMD comparison uses a 440-weight model from 2022, while this work uses a 690,000-weight Allegro-FM fine-tuned model. Even apart from the T2S normalization, the difference in model architecture and the claimed "accuracy-speed trade-off" are not quantified; please state the accuracy difference, e.g., force/energy errors on the same test set.

Circularity Check

1 steps flagged · score 6.0 of 10

The headline 3,780× XS-NNQMD speedup is an arithmetic consequence of the paper's own T2S definition, which divides by neural-network weight count; without that divisor the speedup is ~2.4×.

  1. self definitional [Sec. IV (Current State of the Art), T2S definition for XS-NNQMD; applied in Sec. VII.C.2 and Table II]
    "To account for the accuracy-speed trade-off, we define the T2S of XS-NNQMD as the wall-clock time per MD time step divided by the product of the number of atoms and that of neural-network weights."

    With T2S = t/(N_atoms × W), the ratio of two T2S values equals the raw time-per-atom ratio times W_B/W_A. The paper's own numbers give SOTA t=3142.66 s, N=1.007e12, W=440; this work t=1590.31 s, N=1.2288e12, W=690,000. The raw time-per-atom speedup is only ~2.4×, and the remaining factor is the weight ratio 690,000/440 ≈ 1568, yielding 3780×. Thus the claimed 3,780× improvement is manufactured by including the neural-network weight count in the denominator of the 'time-to-solution' metric, rather than measured as an end-to-end time-to-solution. The definition is chosen after the fact to account for accuracy, but the headline speedup reduces by construction to this normalization choice.

full rationale

The scientific/methodological content of the paper — DC-MESH algorithm, mixed-precision implementation, Allegro-FM XS-NNQMD, scaling measurements — is not circular: the 1.873 EFLOP/s figure, the 152× DC-MESH reduction, and the weak/strong scaling efficiencies are reported measurements from the paper's own runs and prior external benchmarks, not quantities derived by fitting the claims. The only load-bearing circularity is in the XS-NNQMD speedup claim. The paper explicitly defines T2S for XS-NNQMD as wall-clock time divided by (atoms × weights), and then presents the 3,780× ratio as a 'time-to-solution' improvement. That ratio is an identity: it is the raw time-per-atom ratio multiplied by the ratio of the two models' weight counts (440 vs 690,000). Removing the weight normalization reduces the headline to ~2.4×. Because this 3,780× figure is a central Gordon-Bell justification and appears in the abstract, the paper's strongest performance claim is partially circular by definition. The other claims are independent measurements, so the overall circularity is partial rather than total.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central performance claims rest on the validity of the T2S metric, the shadow dynamics approximation, and the sufficiency of mixed precision. The neural network weights and cutoff distance are fitted parameters that underpin the physics capability, though the paper's primary claim is performance, not a physical prediction.

free parameters (2)
  • XS-NNQMD model weights (Allegro-FM fine-tuned) = unknown (690,000 weights)
    The neural network parameters are fitted to training data (Materials Project, SPICE, NAQMD) and are central to the accuracy of the atomic dynamics, though the paper's central claim is performance.
  • Neighbor-list cutoff distance = 5.2 Å
    Used to construct the training dataset for PbTiO3; affects model accuracy but is not justified with a sensitivity study.
assumptions (4)
  • domain assumption Quantum nearsightedness: electronic properties are local, enabling spatial divide-and-conquer.
    Invoked in Sec V.A.2 to justify DC domains; if false, the domain decomposition would lose accuracy.
  • domain assumption Shadow dynamics: the effect of light on nuclear dynamics can be captured by occupation numbers passed from GPU to CPU.
    Sec V.A.3 introduces this approximation to minimize data transfer; its accuracy is not validated in this paper.
  • ad hoc to paper BF16/FP32 mixed precision is sufficient for the nonlocal correction in DC-MESH.
    Sec V.B.7 and VI.C cite ref [34] for this, but do not reproduce the accuracy verification; this is load-bearing for the 1.87 EFLOP/s figure.
  • ad hoc to paper The time-to-solution metric (time per electron or per atom-weight) is a fair way to compare with prior results.
    Sec IV defines T2S and Sec VII.C uses it for the headline speedups; the metric choice materially affects the comparisons.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multiscale light-matter dynamics in quantum materials: from electrons to topological superlattices." pith.science (2026). https://pith.science/paper/4HSFG7SQ

@misc{pith2026250900966,
  author       = {Pith},
  title        = {Pith review of: Multiscale light-matter dynamics in quantum materials: from electrons to topological superlattices},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4HSFG7SQ}},
  note         = {Machine review of arXiv:2509.00966}
}
read the original abstract

Light-matter dynamics in topological quantum materials enables ultralow-power, ultrafast devices. A challenge is simulating multiple field and particle equations for light, electrons, and atoms over vast spatiotemporal scales on Exaflop/s computers with increased heterogeneity and low-precision focus. We present a paradigm shift that solves the multiscale/multiphysics/heterogeneity challenge harnessing hardware heterogeneity and low-precision arithmetic. Divide-conquer-recombine algorithms divide the problem into not only spatial but also physical subproblems of small dynamic ranges and minimal mutual information, which are mapped onto best-characteristics-matching hardware units, while metamodel-space algebra minimizes communication and precision requirements. Using 60,000 GPUs of Aurora, DC-MESH (divide-and-conquer Maxwell-Ehrenfest-surface hopping) and XS-NNQMD (excited-state neural-network quantum molecular dynamics) modules of MLMD (multiscale light-matter dynamics) software were 152- and 3,780-times faster than the state-of-the-art for 15.4 million-electron and 1.23 trillion-atom PbTiO3 material, achieving 1.87 EFLOP/s for the former. This enabled the first study of light-induced switching of topological superlattices for future ferroelectric 'topotronics'.

Figures

Figures reproduced from arXiv: 2509.00966 by the authors.

Figure 1
Figure 1. Software and enabling technologies: MLMD (multiscale light-matter dynamics) simulation & machine-learning software integrates DC-MESH (divide-and-conquer Maxwell-Ehrenfest￾surface hopping) and XS-NNQMD (excited-state neural-network quantum molecular dynamics) modules. DCR (divide-conquer￾recombine) algorithms decompose the problem into not only spatial but also multiphysics subproblems that are mapped onto best char… view at source ↗
Figure 2
Figure 2. (a) Divide-and-conquer domains embedded in a global potential. (b) DC-MESH module consists of (i) LFD to describe light-electron interaction on GPU and (ii) QXMD to describe electron-atom coupling on CPU, with minimal CPU-GPU data transfer via metamodel space algebra (MSA). (c) Machine learning-based XS-NNQMD extends light-matter dynamics to large device-level spatiotemporal scales [PITH_FULL_IMAGE:figures/full_fig… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 12 canonical work pages

  1. [11]

    $ This work Aurora 1.88×10!

    Theta 7.09×10!"$ This work Aurora 1.88×10!"# Table I: State-of-the-art Maxwell-Ehrenfest simulations (subset of the Maxwell-Ehrenfest-surface hopping as in DC-MESH of this work). Work Benchmark system Machine Time-to-solution [sec] PFLOP/s (% of FP64 peak) Qb@ll (2016)

    work page 2016

  2. [1]

    While topological quantum matter was the topic of the 2016 Nobel physics prize, ultrafast control of quantum materials on demand is best achieved via light-matter interaction [3]

    that are best enabled by quantum materials, in which quantum mechanics (QM), such as the topology of electronic wave functions, essentially governs functionality [2]. While topological quantum matter was the topic of the 2016 Nobel physics prize, ultrafast control of quantum materials on demand is best achieved via light-matter interaction [3]. Specifical...

    work page 2016

  3. [2]

    QM/MM can be formulated as extrapolation in a metamodel space, where the two axes are model accuracy and problem size [31, 32, 50]

    akin to multiscale quantum-mechanics/molecular-mechanics (QM/MM) method, which was the topic of the 2013 Nobel chemistry prize to Karplus, Levitt, and Warshel. QM/MM can be formulated as extrapolation in a metamodel space, where the two axes are model accuracy and problem size [31, 32, 50]. The sole assumption is that the difference between QM and MM meth...

    work page 2013

  4. [3]

    Light-matter dynamics is a challenging multiscale problem that encompasses the above multiphysics of vastly different computational characteristics at diverse spatiotemporal scales

    [12]. Light-matter dynamics is a challenging multiscale problem that encompasses the above multiphysics of vastly different computational characteristics at diverse spatiotemporal scales. Its temporal scales extend over fast (10-18 s) elementary processes of light-electron coupling and much slower (10-9 s) materials response through electron-atom coupling...

    work page 2024

  5. [9]

    interconnect is through Xe Link

    Photo-switching of a ferroelectric skyrmion superlattice in PbTiO3. interconnect is through Xe Link. The system interconnect is Slingshot 11, which has high radix 64-port switches and offers adaptive routing, congestion control, and bandwidth guarantees by assigning traffic classes to applications and is coupled with Dragonfly topology. PVC tiles per blad...

    work page 2025

  6. [10]

    We have achieved 1590.31 [sec]/(1,228,800,000,000 [atoms]×690,000 [weights])=1.876×10!%$ [sec] on 10,000 Aurora nodes

    instead to harder excited-state NNQMD. We have achieved 1590.31 [sec]/(1,228,800,000,000 [atoms]×690,000 [weights])=1.876×10!%$ [sec] on 10,000 Aurora nodes. This amounts to 3,780-fold reduction compared to the SOTA T2S of 7.091×10!%# [sec] [11]; see Table II. VIII. IMPLICATIONS As AI becomes ubiquitous at every corner of our society, its enormous power d...

    work page 2024

  7. [21]

    Aluminum, 59,400 electrons IBM BlueGene/Q 8.96×10!% 8.75 (43.5) PWDFT (2020)

    work page 2020

  8. [22]

    Silicon, 3,072 electrons Summit 8.49×10!% 0.12 (2.0) SALMON (2022)

    work page 2022

Show all 13 references
  1. [28]

    V.A.6-7)

    (Sec. V.A.6-7). Implementation innovations complement the algorithmic innovations for exascale performance optimization: (1) Open programming approach based on OpenMP target for portability across exascale computing platforms; (2) data/loop reordering; (3) blocking/tiling; (4)...

    2022

  2. [1687]

    Quantum description of matter was the hallmark of science in the 20th century, but its application was hampered by the exponential computational complexity. For static quantum properties, the complexity was reduced to O(N3) (N is the number of electrons) by density functional ...

    1998

  3. [2005]

    33 Linker, T., et al.: ‘Squishing skyrmions: symmetry guided dynamic transformation’, J Phys Chem Lett, 2022, 13, 11335 34 Piroozan, N., et al.: ‘Impact of varying BLAS precision on DCMESH’, SC-W, 2024, 1468 35 Lam, J.H., et al.: ‘Scalable computation of anisotropic vibrations...

    2022

  4. [2020]

    39 Niklasson, A.M.N., Negre, C.F.A.: ‘Shadow energy functionals and potentials’, J Chem Phys, 2023, 158, 154105 40 Nakano, A., et al.: ‘Massively-parallel algorithms for computational nanoelectronics’, Comput Phys Commun, 1994, 83, 181 41 Richardson, J.L.: ‘Visualizing quantum...

    2023

  5. [2025]

    See Section VI and ALCF cluster details (https://www.alcf.anl.gov/aurora) for workloads and configurations. REFERENCES 1 Ramesh, R., et al.: ‘Roadmap on low-power electronics’, APL Mater, 2024, 12, 099201 2 Tokura, Y., et al.: ‘Emergent functions of quantum materials’, Nat Phy...

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.