Accuracy and scalability of asynchronous compressible flow solver for transitional flows

Aswin Kumar Arumugam; Konduri Aditya; Nagabhushana Rao Vadlamani; Shubham Kumar Goswami

REVIEW 3 major objections 4 minor 34 references

Accuracy and scalability of asynchronous compressible flow solver for transitional flows

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read High-order flow simulations can run on stale boundary data and still capture transition, gaining up to 4x speed at 18,432 cores.

desk verdict Solid integration of asynchrony-tolerant schemes into a production DNS solver with real speedups, but accuracy and speedups are demonstrated at different asynchrony settings — one more airfoil run at L=5 would close the gap. read the letter →

arxiv 2506.03027 v1 pith:WFPJJBRW submitted 2025-06-03 physics.comp-ph physics.flu-dyn

Aswin Kumar Arumugam , Shubham Kumar Goswami , Nagabhushana Rao Vadlamani , Konduri Aditya This is my paper · ORCID

classification physics.comp-phphysics.flu-dyn PACS 47.11.-j

keywords asynchronouscomputingasynchrony-tolerantschemesdirectnumericalsimulationtransitionalflowTaylor-GreenvortexNACA0012airfoilparallelscalabilitycommunicationoverhead

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

The paper's goal is to show that a production-grade, high-order compressible flow solver can tolerate stale, unsynchronized data at subdomain boundaries without losing the accuracy that sensitive flow simulations need. Three test cases — isentropic vortex advection, the Taylor-Green vortex, and separation-induced transition on a NACA0012 airfoil — are used to argue that asynchrony-tolerant (AT) boundary schemes keep the solver's designed second- and fourth-order accuracy and reproduce the synchronous flow physics, while cutting communication overhead enough to deliver speedups of up to about 4x over the synchronous baseline on up to 18,432 cores. Why it matters: communication and synchronization, not computation, are what currently stop direct numerical simulations from scaling to exascale machines, and this paper offers a mathematics-level fix that slots into an existing multi-block solver without changing its interior discretization.

What carries the argument

The asynchrony-tolerant (AT) finite-difference scheme, a boundary stencil that approximates spatial derivatives at PE boundaries from several delayed time levels at buffer points, with coefficients fixed by a Taylor-series constraint system that eliminates lower-order terms so that $q$th-order spatial accuracy survives a delay of up to $L$ time steps. Around this core sit two auxiliary constructions: buffer extrapolation at the zeroth Runge-Kutta stage, and buffer updates in intermediate stages that replace spatial derivatives with temporal derivatives built from the same coefficient-constraint logic, so multistage time integration can skip communication entirely. Two algorithms carry the parallel implementation: the communication-avoiding algorithm (CAA), which communicates periodically and produces uniform delays, and the synchronization-avoiding algorithm (SAA), which initiates non-blocking communication every step without enforcing synchronization and produces stochastic delays.

What would settle it

Run the NACA0012 transition case with CAA-AT at the $L=5$ delay used in the scaling benchmarks and compare the time- and span-averaged surface pressure coefficient and streamwise maximum turbulent kinetic energy against the synchronous baseline; a measurable shift in the separation or transition location near $x/c \approx 0.3$, or the appearance of high-frequency oscillations at PE boundaries, would break the combined accuracy-plus-speedup claim. The paper's error-scaling relation already predicts how boundary error grows with delay, so the test is quantitative.

Watch

Extended reading notes

Core claim

Delayed data at processing-element boundaries does not have to cost accuracy, provided the boundary stencils are rebuilt as asynchrony-tolerant schemes. Naive asynchronous use of standard finite differences collapses to first-order accuracy and emits spurious pressure waves and distorted vortical structures, but the AT-based communication-avoiding and synchronization-avoiding algorithms retain the full design accuracy — second order for LSERK2-CD2-AT2 and fourth order for LSERK4-CD4-AT4, verified up to a maximum allowable delay $L=8$. The same correction keeps sensitive transition physics intact: on the NACA0012 airfoil the asynchronous run matches the synchronous transition location and pressure and turbulent kinetic energy distributions, and in the Taylor-Green vortex it matches the kinetic-energy and enstrophy evolution at a delay where naive asynchronous runs contaminate the whole pressure field.

Load-bearing premise

The flagship claim couples two different settings: the airfoil transition accuracy is checked only with a maximum allowable delay of one time step, while the airfoil speedups are measured with a delay limit of five, so the accuracy of the transitional flow at the delay that produces the reported speedups is assumed rather than demonstrated.

Editorial extensions

If this is right

The asynchronous solver shows clean second- and fourth-order convergence in the isentropic vortex case at maximum allowable delay $L=8$, so a user can relax communication without re-validating the order of accuracy of the boundary treatment.
The error introduced by asynchrony is predictable: for the AT2 scheme it grows as a cubic polynomial in the maximum delay $L$ and linearly in the number of PEs, matching the theoretical relation $\langle\bar{E}\rangle \propto (P/N)\Delta x^q \sum_m \gamma_m \bar{k}^m$, so the delay $L$ becomes a quantity a user can budget against accuracy.
Strong-scaling plateaus move to much larger core counts: the synchronous solver's execution time plateaus between 2048 and 8192 cores while CAA-AT and SAA-AT stay near ideal scaling, with communication time, not compute, dominating the synchronous cost; speedups reach 4.26x at 8192 cores on the Taylor-Green case, about 3x on the vortex case, and about 2.7x on the airfoil case.
For sensitive flows the AT correction is decisive: CAA-AS, which uses standard schemes with delayed data, distorts vortical structures and radiates spurious pressure waves in the Taylor-Green vortex, whereas CAA-AT matches the synchronous solution and captures the key vortical structure across the PE boundary.
The synchronization-avoiding variant has architecture-dependent, stochastic delays, and the paper shows its delay distribution — mostly one or two time steps, with larger delays becoming more probable at higher PE counts — correlates directly with deviations from ideal scaling on the two test supercomputers.

Reading between the lines

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

The accuracy-and-speedup promise for the most sensitive case couples two settings the paper never combines: airfoil transition accuracy is demonstrated at maximum delay $L=1$, while the airfoil speedup benchmarks run at $L=5$. The natural next experiment is the airfoil physics at $L=5$, and the paper's own error-scaling relation makes the expected accuracy penalty quantitative rather than speculat
Because AT corrections are localized at boundaries and arithmetically dense, their relative cost should shrink as subdomains shrink; the measured speedups are therefore likely to grow, not shrink, as core counts pass 18,432, provided the delay distribution stays thin-tailed.
The same machinery is a generic pattern for stencil-based exascale PDE solvers — the authors cite discontinuous-Galerkin and reacting-flow extensions of the approach — so 'rebuild boundary stencils in time and skip synchronization' may transfer well beyond structured finite-difference solvers.
The validated error-scaling law turns the maximum delay $L$ into a tunable parameter: an application can set the delay from an accuracy budget up front, which is what would make asynchronous execution a rigorous option for transition-sensitive engineering simulations rather than an uncontrolled approximation.

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.

Desk Editor's Note

read the letter

Two things to know. First, this is a genuine integration result: the asynchrony-tolerant (AT) schemes from the authors' earlier papers are now inside COMP-SQUARE, a production multi-block compressible DNS solver, and the paper shows them working on the isentropic vortex, the Taylor-Green vortex, and — the new piece — a NACA0012 boundary layer with separation-induced transition. Second, the accuracy evidence and the speedup evidence are taken at different asynchrony severities, and that gap sits right on the central claim.

Credit where it is due. The airfoil case is a real step beyond the canonical PDE tests in the prior work; seeing cp and TKE_max profiles agree with the synchronous baseline on a sensitive transition is exactly what the method needed to show. The accuracy verification is careful: order convergence to the expected second and fourth order, the error-scaling law (cubic in L, linear in P) confirmed against fits, and energy spectra showing that AT removes the high-wavenumber contamination that plain delayed schemes produce. The measured delay distributions for SAA on two architectures are a useful addition. The AT coefficients are given in full, so the accuracy tests are reproducible from the paper alone.

The soft spots, in proportion. The mismatch the report flags is real. The airfoil physics comparison runs CAA-AT at L=1 — essentially only the intra-RK-stage asynchrony — while the 2.47x and 2.73x speedups are measured at L=5. The TGV shows the same pattern: enstrophy and kinetic energy match at L=5, scaling at L=10. Their own error law makes this matter: the boundary-error constant grows as a degree-C polynomial in the delay, C=5 for AT4, so the error budget at the speedup settings is not the one that was tested. I would soften the stress-test's 'three orders of magnitude' — that is the polynomial constant, not the actual error, which is scaled by (P/N) times the spatial step to the q-th power — but the qualitative point stands: the combined claim of accurate transition plus speedup is verified piecewise, not together. A referee should ask for one airfoil run at L=5 with a quantitative comparison, such as transition location or an error norm. The synchronous baseline also blocks on MPI_Wait with no communication overlap — the paper acknowledges overlap as a possible optimization — so the ~4x speedup is against a conservative baseline, though this is consistent with the asynchronous-CFD literature. No code or data are released, only 'available upon reasonable request.'

Bottom line: the central idea holds up. This is a solid engineering-science paper for CFD people worried about exascale communication costs, and it deserves serious review. The missing L=5 airfoil accuracy run is an addressable revision, not a fatal flaw.

Referee Report

3 major / 4 minor

Summary. The paper integrates asynchrony-tolerant (AT) schemes into the COMP-SQUARE compressible Navier-Stokes solver via two parallel algorithms: a communication-avoiding algorithm (CAA) and a synchronization-avoiding algorithm (SAA). It reports order-of-accuracy verification for isentropic vortex advection, accuracy comparisons for the Taylor-Green vortex (TGV) and for transitional flow over a NACA0012 airfoil, and strong/weak scaling experiments on up to 18,432 cores with speedups up to about 4x relative to the synchronous baseline. The abstract and conclusions claim that the AT-based asynchronous solver preserves high-order accuracy for sensitive transitional flows while substantially improving scalability.

Significance. If the results hold, the paper demonstrates a practically important step: prior AT scheme theory, developed on canonical PDEs, can be embedded in an established multi-block structured CFD solver and produce accurate transitional-flow predictions while reducing communication overhead. The authors deserve credit for verifying the nominal convergence orders of CAA-AT and SAA-AT, for matching the TGV evolution to a pseudo-spectral reference at L=5, and for quantifying SAA delay statistics and correlating them with performance. The main weakness is that accuracy and scalability are demonstrated at different asynchrony settings, so the paper's combined claim "accurate for transitional flows at the settings that produce the speedups" is not yet directly supported. The AT scheme derivations are imported from prior work rather than derived here, but this is not a circularity problem because the numerical behavior is validated against independent references.

major comments (3)

[Sec. 5.3, Figs. 9 and 11] The airfoil accuracy comparison is performed with CAA-AT at L=1 (Fig. 9 caption), while the reported speedups of 2.47x and 2.73x are measured at L=5 (Fig. 11 and surrounding text). Under the intended CAA semantics, L=1 bounds the time-step delay to at most one step, so the multi-step delayed-data regime that produces the communication savings is not exercised in the accuracy test. Since the paper's own error law (Sec. 5.1, Fig. 4a) shows boundary error growing as a polynomial in the delay, with degree related to the number of AT time levels (C=5 for AT4), the effect of L=5 on the separation-induced transition cannot be inferred from the L=1 result. Please add a quantitative accuracy comparison at L=5 (e.g., transition location, cp error, and TKEmax error norms) or explicitly restrict the accuracy claim to L=1.
[Sec. 5.2, Figs. 6 and 8] The TGV accuracy comparison is shown at L=5 for CAA-AT (Fig. 6 caption), while the strong- and weak-scaling speedups, including the headline 4.26x at 8192 cores, are reported at L=10 (Fig. 8 and text). The kinetic-energy and enstrophy agreement in Fig. 6 therefore does not cover the asynchrony level used in the scaling benchmarks. Please provide an accuracy-versus-L study or a TGV accuracy comparison at L=10, or report the scaling results at the same L used in the accuracy tests.
[Sec. 4.2, Algorithm 2] The CAA pseudocode is internally inconsistent with the reported L values. As printed, the conditions "0 <= n%L <= C-1" and "n%L == C-1" mean that for L=1 and C=5 the communication/reset branch is never taken (since n%1=0 and 0 != 4), so the delay would never reset; for L=10 and C=5 the reset would occur at n%L=4, giving a maximum delay of about 5 rather than 10. This contradicts the stated definition of L as the maximum allowable delay and the reported L=1 and L=10 experiments. Please correct the pseudocode (e.g., replace C-1 with L-1 or clarify the modulo convention) and describe how the L=1 case is actually implemented and verified.

minor comments (4)

[Sec. 5.1, Fig. 4] The dashed curve in Fig. 4(a) is described as a best-fit cubic polynomial, but the fitted polynomial is not reported; giving the fit coefficients would allow readers to check whether the L-dependence is indeed a degree-(C-1) polynomial rather than just a visual fit.
[Sec. 5.1, Fig. 3] In Fig. 3(b), the legend lists both CAA-AT and SAA-AT, but the two curves may overlap; please use distinguishable markers or explicitly state where they coincide.
[Sec. 5.3, Fig. 10] The caption describes panels (c) and (d) as comparisons against Ref. [33], but the text does not state the quantitative difference in transition location between SA and CAA-AT; a numerical value would strengthen the claim of consistent transition prediction.
[Eq. 16] The pressure relation is written as "p_inf - p = ...", which is unconventional; writing "p = p_inf - ..." would make the sign convention clearer.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: accuracy claims are validated against external references, and the L=1 vs L=5 speedup mismatch is a coverage gap, not a circular reduction.

full rationale

The paper's claimed derivation chain is not circular. The asynchrony-tolerant (AT) schemes are imported from prior published work [10,13] as explicit coefficient constraints (Eqs. 6 and 13) with stated assumptions (CFL exponent r, target order q), and they are not defined in terms of the flow solutions they are used to predict. The core accuracy claims are independently validated: order-of-accuracy via convergence against the exact isentropic vortex solution (Fig. 3a), Taylor-Green vortex accuracy against the external pseudo-spectral reference [32] (Fig. 6), and airfoil transition against the high-fidelity LES of Visbal and Garmann [33] (Fig. 10c). The error-scaling relation from Ref. [10] is invoked to predict a cubic-in-L and linear-in-P dependence, which the paper then fits and compares; because the functional form is derived rather than fitted and is not used to define the solver's error, this is validation, not a fitted-input-as-prediction. The L=1 accuracy vs L=5 speedup mismatch in Sec. 5.3 (and the analogous L=5 vs L=10 gap in the TGV scaling study) is a coverage or verification gap, not a circularity: it means the combined accuracy-plus-speedup claim is not demonstrated at the exact speedup settings, but no equation in the paper makes a predicted quantity equal to an input by construction. Self-citations are present but not load-bearing: [10,13] supply parameter-free coefficient constraints with stated assumptions that do not include the target results, and the paper's own numerical tests provide independent evidence for the central claims.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central accuracy claims rest on the imported asynchrony-tolerant scheme framework and on assumptions about delayed boundary data, filtering, and solver stability, not on new fitted constants. Delay, time-step, and filter settings are chosen by hand and differ between accuracy and scaling runs. No new physical entities are introduced.

free parameters (3)

Maximum allowable delay L = 1 (airfoil accuracy), 5 (airfoil scaling and TGV accuracy), 8 (vortex accuracy), 10 (TGV scaling)
User-chosen parameter controlling how stale boundary data may be; the accuracy and scaling demonstrations use different L values for the airfoil, which weakens the coupling between accuracy and speedup.
Time step size = 2e-3 and 1e-3 for vortex SA and asynchronous runs; 1e-3 and 1e-4 for TGV SA and asynchronous runs; airfoil not stated
Asynchronous runs use smaller time steps to satisfy stricter CFL limits, so accuracy and cost comparisons are not made at identical time step sizes.
Filter coefficient alpha_f = 0.499
Tenth-order implicit filter coefficient used for the Taylor-Green vortex; its interaction with boundary delay errors is not isolated.

assumptions (4)

domain assumption AT Taylor-coefficient constraints (Eq. 6 in Sec. 3.1) carry over from scalar PDEs to the nonlinear compressible Navier-Stokes equations on curvilinear grids.
The paper imports AT schemes from prior work [10] and applies them to full flux functions without new local truncation analysis for nonlinear, metric, or filter terms.
domain assumption Delayed PE-boundary data can be modeled by fixed integer delays k in Taylor expansions, with random delays distributed around a bounded mean.
Central to Sec. 3.1 and to the error scaling check in Sec. 5.1.
domain assumption High-order Pade filtering suppresses asynchrony-induced high-frequency oscillations without altering physical flow structures.
Used in Sec. 2 and the TGV test; no study isolates filter strength versus delay error.
domain assumption COMP-SQUARE's standard high-order finite difference and compact schemes are stable and accurate on the tested curvilinear multi-block grids.
Assumed from the solver's prior validation, cited in Sec. 1 through references [18-24].

how reviews work

0 comments

Cite this review

Pith. "Pith review of Accuracy and scalability of asynchronous compressible flow solver for transitional flows." pith.science (2026). https://pith.science/paper/WFPJJBRW

@misc{pith2026250603027,
  author       = {Pith},
  title        = {Pith review of: Accuracy and scalability of asynchronous compressible flow solver for transitional flows},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WFPJJBRW}},
  note         = {Machine review of arXiv:2506.03027}
}

abstract

To overcome the communication bottlenecks observed in state-of-the-art parallel time-dependent flow solvers at extreme scales, an asynchronous computing approach that relaxes communication and synchronization at a mathematical level was previously developed. This approach preserves high-order accuracy of computations near processing element boundaries using asynchrony-tolerant (AT) schemes while significantly improving the scalability. The numerical properties of the AT schemes were studied based on simple linear and nonlinear partial differential equations (PDEs) in previous works. Allowing asynchrony in numerical schemes can minimize communication overheads in a parallel setting in two ways: one that avoids communication over a few predetermined time steps, and the other that initiates communications without enforcing synchronization. In this study, the asynchronous algorithms are incorporated into the high-order compressible flow solver COMP-SQUARE, which solves practically relevant flow problems in complex geometries in a multi-block framework. The numerical efficacy and scalability of the two asynchronous algorithms are demonstrated for three test cases: isentropic advection of a vortex, the Taylor-Green vortex, and a much more sensitive case of the flow transitioning on a NACA0012 airfoil. Speed-ups of up to $4\times$ with respect to the baseline synchronous algorithm are observed in the scaling experiments performed on up to 18,432 cores. The results of this study demonstrate the applicability of AT schemes on established CFD solvers to improve scalability at extreme scales as the scientific computing environment moves to the exascale era.

Figures

Figures reproduced from arXiv: 2506.03027 by the authors.

**Figure 1.** (a) An illustration of communications with relaxed synchronization at the interface between two processing elements (PEs). The delay in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** (a) Vorticity magnitude contours of initial flow field. Instantaneous contours of errors in vorticity magnitude after one advection cycle [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗

**Figure 3.** (a) Order convergence plot for density error for all the cases in Table 1, and (b) energy spectra of [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

Figures from the paper (9 more)

**Figure 4.** Figure 4: Plots of error in u-velocity after t = 4 for the isentropic vortex advection solved on a 20482 grid against (a) maximum allowable delay L on 1024 PEs and (b) number of PEs P with L = 4 for CAA-AT corresponding to case 3 in [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

**Figure 5.** Figure 5: Strong scaling results of isentropic vortex advection on a [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]

**Figure 6.** Figure 6: Time evolution of volume-averaged (a) kinetic energy and (b) enstrophy on [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

**Figure 7.** Figure 7: (a) Global contours of pressure, and (b) magnified contours of vorticity magnitude around a vortical structure at the [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

**Figure 8.** Figure 8: (a) Strong scaling results for resolutions of [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]

**Figure 9.** Figure 9: Instantaneous iso-surfaces of Q = 10 for the flow around NACA0012 airfoil at AoA 4 ◦ with (a) SA and (b) CAA-AT (cases 4 and 6, respectively, in [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]

**Figure 10.** Figure 10: Contours of (a) pressure coefficient (cp) and (b) turbulent kinetic energy (TKE, magnified near the suction surface) obtained from time and span averaged fields. The colored contours in the background and the overlaid dashed lines correspond to SA and CAA-AT, respecti…

**Figure 11.** Figure 11: Strong scaling graphs for two resolutions of the airfoil case on (a) PARAM Pravega and (b) PARAM Rudra supercomputers, for SA, [PITH_FULL_IMAGE:figures/full_fig_p017_11.png]

**Figure 12.** Figure 12: Probability distribution of delays for the airfoil case on (a) PARAM Pravega and (b) PARAM Rudra supercomputers, for SAA-AT (see [PITH_FULL_IMAGE:figures/full_fig_p018_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

[1]

J. H. Chen, A. Choudhary, B. D. Supinski, M. Devries, E. R. Hawkes, S. Klasky, W. K. Liao, K. L. Ma, J. Mellor-Crummey, N. Podhorszki, R. Sankaran, S. Shende, C. S. Yoo, Terascale direct numerical simulations of turbulent combustion using S3D, Computational Science and Discovery 2 (2009)

work page 2009
[2]

K. W. Jenkins, R. S. Cant, Direct numerical simulation of turbulent flame kernels, in: D. Knight, L. Sakell (Eds.), Recent Advances in DNS and LES, Springer Netherlands, Dordrecht, 1999, pp. 191–202

work page 1999
[3]

Desjardins, G

O. Desjardins, G. Blanquart, G. Balarac, H. Pitsch, High order conservative finite difference scheme for variable density low Mach number turbulent flows, Journal of Computational Physics 227 (2008) 7125–7159

work page 2008
[4]

T. C. Germann, Co-design in the exascale computing project, The International Journal of High Performance Computing Applications 35 (2021) 503–507

work page 2021
[5]

Amitai, S

D. Amitai, S. Itzikowitz, Parallel adaptive and time-stabilizing schemes for constant-coefficient parabolic PDE’s, 1992. 20 Derivative Expression (∂ϕ/∂t) (0) i 1 12∆t 2˜k3 +15 ˜k2 +35 ˜k+25 ϕn−˜k i − 1 6∆t 4˜k3 +27 ˜k2 +52 ˜k+24 ϕn−˜k−1 i + 1 2∆t 2˜k3 +12 ˜k2 +19 ˜k+6 ϕn−˜k−2 i − 1 6∆t 4˜k3 +21 ˜k2 +28 ˜k+8 ϕn−˜k−3 i + 1 12∆t 2˜k3 +9 ˜k2 +11 ˜k+3 ϕn−˜k−4 ...

work page 1992
[6]

Amitai, A

D. Amitai, A. Averbuch, S. Itzikowitz, E. Turkel, Asynchronous and corrected-asynchronous finite difference solutions of PDEs on MIMD multiprocessors, 1994

work page 1994
[7]

Konduri, D

A. Konduri, D. A. Donzis, Poster: Asynchronous Computing for Partial Differential Equations at Extreme Scales, in: 2012 SC Companion: High Performance Computing, Networking Storage and Analysis, 2012, pp. 1444–1444

work page 2012
[8]

D. A. Donzis, K. Aditya, Asynchronous finite-difference schemes for partial differential equations, Journal of Computational Physics 274 (2014) 370–392

work page 2014

Show all 34 references

[9]

Mittal, S

A. Mittal, S. Girimaji, Proxy-equation paradigm: A strategy for massively parallel asynchronous computations, Physical Review E 96 (2017)

2017
[10]

Aditya, D

K. Aditya, D. A. Donzis, High-order asynchrony-tolerant finite difference schemes for partial differential equations, Journal of Computational 21 Physics 350 (2017) 550–572

2017
[11]

Aditya, T

K. Aditya, T. Gysi, G. Kwasniewski, T. Hoefler, D. A. Donzis, J. H. Chen, A scalable weakly-synchronous algorithm for solving partial differential equations, 2019

2019
[12]

Kumari, D

K. Kumari, D. A. Donzis, Direct numerical simulations of turbulent flows using high-order asynchrony-tolerant schemes: Accuracy and performance, Journal of Computational Physics 419 (2020)

2020
[13]

S. K. Goswami, V . J. Matthew, K. Aditya, Implementation of low-storage Runge-Kutta time integration schemes in scalable asynchronous partial differential equation solvers, Journal of Computational Physics 477 (2023)

2023
[14]

Kumari, E

K. Kumari, E. Cleary, S. Desai, D. A. Donzis, J. H. Chen, K. Aditya, Evaluation of finite difference based asynchronous partial differential equations solver for reacting flows, Journal of Computational Physics 477 (2023)

2023
[15]

S. K. Goswami, K. Aditya, An asynchronous discontinuous-Galerkin method for solving PDEs at extreme scales, 2022

2022
[16]

S. K. Goswami, K. Aditya, An asynchronous discontinuous Galerkin method for massively parallel PDE solvers, Computer Methods in Applied Mechanics and Engineering 430 (2024)

2024
[17]

A. K. Arumugam, K. Aditya, An asynchronous discontinuous Galerkin method for combustion simulations, 2025

2025
[18]

N. R. Vadlamani, P. G. Tucker, P. Durbin, Distributed Roughness Effects on Transitional and Turbulent Boundary Layers, Flow, Turbulence and Combustion 100 (2018) 627–649

2018
[19]

A. S. Malathi, M. Nardini, A. Vaid, N. R. Vadlamani, R. D. Sandberg, Profile Loss Reduction of High-Lift Turbine Blades with Rough and Ribbed Surfaces, Journal of Turbomachinery 145 (2023)

2023
[20]

N. R. Vadlamani, P. G. Tucker, Eddy resolving simulations of intake under crosswinds, in: M. V . Salvetti, V . Armenio, J. Fr ¨ohlich, B. J. Geurts, H. Kuerten (Eds.), Direct and Large-Eddy Simulation XI, Springer International Publishing, Cham, 2019, pp. 523–529

2019
[21]

Tyacke, N

J. Tyacke, N. R. Vadlamani, W. Trojak, R. Watson, Y . Ma, P. G. Tucker, Turbomachinery simulation challenges and the future, 2019

2019
[22]

A. I. Arasu, S. S. Sarvankar, N. R. Vadlamani, Distortion Control in Intakes Subject to Strong Crosswinds Using Steady V ortex Generator Jets, AIAA Journal (2024) 1–16

2024
[23]

Y . Lin, R. Vadlamani, M. Savill, P. Tucker, Wall-resolved large eddy simulation for aeroengine aeroacoustic investigation, in: Aeronautical Journal, volume 121, Cambridge University Press, 2017, pp. 1032–1050

2017
[24]

Nampelly, A

G. Nampelly, A. S. Malathi, A. Vaid, N. R. Vadlamani, S. Rengarajan, K. Kontis, Surface Roughness Effects on Cavity Flows, Flow, Turbulence and Combustion 109 (2022) 1215–1239

2022
[25]

S. K. Lele, Compact Finite Difference Schemes with Spectral-like Resolution, Journal of Computational Physics 103 (1992) 42

1992
[26]

M. R. Visbal, D. V . Gaitonde, On the use of higher-order finite-difference schemes on curvilinear and deforming meshes, Journal of Computational Physics 181 (2002) 155–185

2002
[27]

J. H. Williamson, Low-storage Runge-Kutta schemes, Journal of Computational Physics (1980) 3548–56

1980
[28]

C. A. Kennedy, M. H. Carpenter, R. M. Lewis, Low-storage, explicit Runge-Kutta schemes for the compressible Navier-Stokes equations, Applied Numerical Mathematics 35 (2000) 177–219

2000
[29]

M. H. Carpenter, C. A. Kennedy, Fourth-order 2N-storage Runge-Kutta schemes, Technical Report, NASA Technical Memorandum 109112, 1994

1994
[30]

L. T. Diosady, S. M. Murman, Case 3.3: Taylor-Green V ortex Evolution, Technical Report, NASA Ames Research Center, 2015

2015
[31]

DeBonis, Solutions of the Taylor-Green V ortex Problem Using High-Resolution Explicit Finite Difference Methods, 2013

J. DeBonis, Solutions of the Taylor-Green V ortex Problem Using High-Resolution Explicit Finite Difference Methods, 2013

2013
[32]

W. M. van Rees, A. Leonard, D. I. Pullin, P. Koumoutsakos, A comparison of vortex and pseudo-spectral methods for the simulation of periodic vortical flows at high Reynolds numbers, Journal of Computational Physics 230 (2011) 2794–2805

2011
[33]

M. R. Visbal, D. J. Garmann, Analysis of dynamic stall on a pitching airfoil using high-fidelity large-eddy simulations, in: AIAA Journal, volume 56, American Institute of Aeronautics and Astronautics Inc., 2018, pp. 46–63

2018
[34]

D. V . Gaitonde, M. R. Visbal, High-Order Schemes for Navier-Stokes Equations: Algorithm and Implementation Into FDL3DI, 1998. 22

1998

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

J. H. Chen, A. Choudhary, B. D. Supinski, M. Devries, E. R. Hawkes, S. Klasky, W. K. Liao, K. L. Ma, J. Mellor-Crummey, N. Podhorszki, R. Sankaran, S. Shende, C. S. Yoo, Terascale direct numerical simulations of turbulent combustion using S3D, Computational Science and Discovery 2 (2009)

work page 2009

[2] [2]

K. W. Jenkins, R. S. Cant, Direct numerical simulation of turbulent flame kernels, in: D. Knight, L. Sakell (Eds.), Recent Advances in DNS and LES, Springer Netherlands, Dordrecht, 1999, pp. 191–202

work page 1999

[3] [3]

Desjardins, G

O. Desjardins, G. Blanquart, G. Balarac, H. Pitsch, High order conservative finite difference scheme for variable density low Mach number turbulent flows, Journal of Computational Physics 227 (2008) 7125–7159

work page 2008

[4] [4]

T. C. Germann, Co-design in the exascale computing project, The International Journal of High Performance Computing Applications 35 (2021) 503–507

work page 2021

[5] [5]

Amitai, S

D. Amitai, S. Itzikowitz, Parallel adaptive and time-stabilizing schemes for constant-coefficient parabolic PDE’s, 1992. 20 Derivative Expression (∂ϕ/∂t) (0) i 1 12∆t 2˜k3 +15 ˜k2 +35 ˜k+25 ϕn−˜k i − 1 6∆t 4˜k3 +27 ˜k2 +52 ˜k+24 ϕn−˜k−1 i + 1 2∆t 2˜k3 +12 ˜k2 +19 ˜k+6 ϕn−˜k−2 i − 1 6∆t 4˜k3 +21 ˜k2 +28 ˜k+8 ϕn−˜k−3 i + 1 12∆t 2˜k3 +9 ˜k2 +11 ˜k+3 ϕn−˜k−4 ...

work page 1992

[6] [6]

Amitai, A

D. Amitai, A. Averbuch, S. Itzikowitz, E. Turkel, Asynchronous and corrected-asynchronous finite difference solutions of PDEs on MIMD multiprocessors, 1994

work page 1994

[7] [7]

Konduri, D

A. Konduri, D. A. Donzis, Poster: Asynchronous Computing for Partial Differential Equations at Extreme Scales, in: 2012 SC Companion: High Performance Computing, Networking Storage and Analysis, 2012, pp. 1444–1444

work page 2012

[8] [8]

D. A. Donzis, K. Aditya, Asynchronous finite-difference schemes for partial differential equations, Journal of Computational Physics 274 (2014) 370–392

work page 2014