A Pre-Dispatch Resonance Safety Criterion for AI Training Clusters

Abanish Tiwari; Chandan Chaudhary; Joydeep Mitra; Mohammed Ben-Idris; Yansong Pei

arxiv: 2606.22096 · v1 · pith:6B4PEE77new · submitted 2026-06-20 · 📡 eess.SY · cs.SY

A Pre-Dispatch Resonance Safety Criterion for AI Training Clusters

Chandan Chaudhary , Abanish Tiwari , Yansong Pei , Mohammed Ben-Idris , Joydeep Mitra This is my paper

Pith reviewed 2026-06-26 11:43 UTC · model grok-4.3

classification 📡 eess.SY cs.SY

keywords AI training clustersgrid resonanceiteration periodtwo-area swing modelpower swingpre-dispatch criterionelectromechanical modesGPU cluster

0 comments

The pith

A closed-form criterion derived from swing equations bounds the maximum size of AI training clusters to prevent resonance with grid modes.

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

Hyperscale AI training clusters create periodic power swings that can resonate with electromechanical modes on the transmission grid when the training iteration period falls in the one-to-ten-second range. The paper inverts the steady-state forced two-area swing equations to produce a closed-form criterion that gives the largest safe cluster size for any proposed iteration period. This criterion identifies danger bands around resonant periods and accounts for square-wave harmonics using parameters from planning studies. If the criterion holds, grid operators gain an analytic tool to screen large loads before dispatch and can treat the iteration period as an adjustable safety parameter. Application to the IEEE 39-bus test system illustrates that clusters up to 66,900 GPUs remain safe at resonance under light damping, while shifting the schedule by less than one second reduces deviation by a factor of 7.4.

Core claim

The paper derives a closed-form pre-dispatch safety criterion by inverting the steady-state forced two-area swing equations. The criterion bounds the maximum cluster size a grid can absorb at any iteration period, defines danger bands, extends to square-wave harmonics, and parameterizes the response from eigenanalysis and GPU specifications. When applied to the IEEE 39-bus system at a representative duty cycle, it yields a maximum safe cluster of 66,900 GPUs at resonance under light damping, with rescheduling away from resonance reducing deviation by 7.4 times.

What carries the argument

The inverted steady-state forced two-area swing equations, which map cluster power amplitude and modal damping to steady-state angle or voltage deviation bounds.

If this is right

Iteration period selection becomes a grid-safety control variable.
Small changes in schedule timing can substantially reduce mode excitation.
The criterion supplies an analytic screening method for large periodic loads.
Harmonic components of the square wave must be checked in addition to the fundamental.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Job schedulers in AI clusters could incorporate grid frequency data to avoid resonant periods.
The approach may extend to other synchronous large loads such as cryptocurrency mining farms.
Validation against full-scale interconnection models would strengthen the two-area approximation.

Load-bearing premise

The aggregate power draw of the training cluster follows an exact square wave at the iteration period, and the two-area model with planning-study parameters adequately captures the relevant inter-area modes.

What would settle it

Direct measurement of power consumption waveform from an operating hyperscale training cluster or field recording of inter-area mode response exceeding the predicted steady-state deviation for a known cluster size.

Figures

Figures reproduced from arXiv: 2606.22096 by Abanish Tiwari, Chandan Chaudhary, Joydeep Mitra, Mohammed Ben-Idris, Yansong Pei.

**Figure 2.** Figure 2: summarizes the screening workflow. The criterion compares the planned cluster size against the safe bound and returns either a dispatch clearance or a schedule adjustment. Job specification NGPU, Titer, d; PTDP, Pidle, ηPS Planning-study modal data fk, ζk, E1, E2 ⇒ Eeq, B Start: new training job request BSP forcing model ∆P1 = NGPU a(d) at fc = 1/Titer Criterion (8) evaluate N∗ (Titer); safe if NGPU ≤ N∗ I… view at source ↗

**Figure 3.** Figure 3: Pre-dispatch criterion for the inter-area mode ( [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Absolute frequency deviation at area-1 buses for [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: Resonant-floor contours N∗ min (thousands of GPUs) over the (fk, ζk) plane at the 39-bus kinetic energy. Interconnection lie at 0.16–0.32 Hz [21], placing resonance periods at 3.1–6.3 s, which is the middle of the production scheduling window. Modes in that band impose floors roughly three times tighter than the 0.6 Hz mode studied here, so at interconnection scale resonance screening is not a niche check … view at source ↗

read the original abstract

Hyperscale AI training clusters operate under the Bulk Synchronous Parallel protocol, which impose a periodic power swing on the transmission grid. Every GPU in the job transitions between compute and idle in lockstep, so the aggregate power traces a square wave at the training iteration period. Production iteration periods of one to ten seconds place the forcing frequency within the inter-area electromechanical mode band of large interconnections, where a training schedule can drive a mode at resonance. This paper derives a closed-form pre-dispatch safety criterion that bounds the maximum cluster size a grid can absorb at any proposed iteration period. The derivation inverts the steady-state forced two-area swing equations. The criterion defines a danger band of iteration periods, extends to the square-wave harmonics, and parameterizes the modal response from planning-study eigenanalysis and the forcing amplitude from GPU specifications. Applied to the IEEE 39-bus system at a production-representative duty cycle, the criterion shows that the maximum safe cluster at resonance is $66\,900$ GPUs under light damping. Rescheduling the same job less than one second away from resonance reduces the deviation $7.4\times$ with no hardware change. These results establish the training iteration period as a controllable grid-safety parameter and supply the analytic screening tool that reliability directives on current large loads lack.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a closed-form pre-dispatch bound on AI cluster size to avoid grid resonance by inverting two-area swing equations for square-wave forcing, but the reduced model's ability to represent real inter-area response is the main question.

read the letter

The main takeaway is a new closed-form criterion that bounds the maximum number of GPUs in an AI training cluster for a proposed iteration period so that the square-wave power swing does not hit resonance with inter-area modes. It does this by inverting the steady-state forced response of the two-area swing equations.

What is new is the specific application to the bulk synchronous parallel protocol and the extension to harmonics. The paper does well in showing how to parameterize it from existing eigenanalysis and hardware data, and the example on the IEEE 39-bus system demonstrates both the danger band and the large benefit from small schedule adjustments without changing hardware.

The soft spots are in the model and validation. The two-area swing model plus planning eigenanalysis may not represent the actual inter-area response to a localized periodic load, especially since real systems have multiple modes whose observability depends on electrical distance. The paper's numerical results are on a standard test system, but without comparison to a full-order model or real measurements, it's hard to know if the 66,900 GPU limit or the 7.4 times reduction holds up. The assumption that the aggregate power exactly follows a square wave at the iteration period is also an idealization that could be off in practice.

This paper is for people in power system operations and planning who are starting to deal with very large computing loads, and for AI infrastructure teams who can control job schedules. A reader looking for an analytic screening tool rather than full dynamic simulation would get something from it. It deserves a serious referee because the interface between hyperscale loads and grid stability is important and the approach is direct, even if the reduced model needs more scrutiny.

I would send it to peer review with a request for additional validation of the forced response.

Referee Report

2 major / 2 minor

Summary. The paper derives a closed-form pre-dispatch safety criterion bounding the maximum size of an AI training cluster (under Bulk Synchronous Parallel scheduling) that a grid can absorb without driving inter-area electromechanical modes into resonance. The derivation inverts the steady-state forced response of the two-area swing equations, with modal parameters taken from planning-study eigenanalysis and forcing amplitude from GPU power specifications; the criterion identifies a danger band of iteration periods (one to ten seconds), extends to square-wave harmonics, and is applied to the IEEE 39-bus system to obtain a maximum safe cluster of 66,900 GPUs at resonance under light damping, with a 7.4× reduction in deviation obtained by rescheduling the iteration period less than one second away from resonance.

Significance. If the reduced-order model is shown to be representative, the work supplies an analytic screening tool that treats the training iteration period as a controllable grid-safety parameter, addressing a gap in current reliability directives for large periodic loads. The closed-form nature and use of standard test systems and externally sourced modal data are strengths that could enable rapid pre-dispatch screening.

major comments (2)

[Abstract / derivation] Abstract and derivation: the central bound (66,900 GPUs) and 7.4× rescheduling claim rest on the two-area swing model plus planning-study eigenanalysis being sufficient to predict forced-oscillation amplitude at the cluster bus; the IEEE 39-bus system is a multi-machine network whose inter-area modes have location-dependent participation factors and observability that a two-area reduction cannot capture, so the criterion may not bound the actual response when the cluster is electrically distant from the mode shape.
[Abstract] Abstract: the forcing is taken as an ideal square wave whose amplitude is set directly from GPU specifications; no error analysis or sensitivity to deviations from perfect square-wave behavior (e.g., stochastic compute/idle timing jitter within an iteration) is supplied, yet this directly scales the steady-state response amplitude used to obtain the safety bound.

minor comments (2)

[Abstract] The abstract states that a derivation exists and reports numerical results but does not display the inverted closed-form expression or the definition of the danger band; including the key equation would improve readability.
No table or figure caption clarifies how the 7.4× factor is computed (e.g., which damping ratio, which harmonic, exact frequency offset); a small table of sensitivity cases would help.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for these detailed comments on the modeling assumptions. We respond to each major comment below, indicating planned revisions where appropriate.

read point-by-point responses

Referee: [Abstract / derivation] Abstract and derivation: the central bound (66,900 GPUs) and 7.4× rescheduling claim rest on the two-area swing model plus planning-study eigenanalysis being sufficient to predict forced-oscillation amplitude at the cluster bus; the IEEE 39-bus system is a multi-machine network whose inter-area modes have location-dependent participation factors and observability that a two-area reduction cannot capture, so the criterion may not bound the actual response when the cluster is electrically distant from the mode shape.

Authors: The two-area swing equations are inverted to obtain the closed-form safety criterion, which is a deliberate modeling choice to yield an analytic pre-dispatch tool rather than a full-order simulation. Modal parameters (frequency, damping ratio, and mode shape) are taken directly from eigenanalysis of the complete IEEE 39-bus system, so the forcing amplitude is scaled by the participation at the cluster bus. We agree that a two-area reduction does not fully capture all location-dependent observability effects in a multi-machine network. We will revise the manuscript to state this limitation explicitly, emphasize that the criterion is intended as a conservative screening bound, and note that full dynamic simulation remains necessary for final validation at a specific bus. revision: partial
Referee: [Abstract] Abstract: the forcing is taken as an ideal square wave whose amplitude is set directly from GPU specifications; no error analysis or sensitivity to deviations from perfect square-wave behavior (e.g., stochastic compute/idle timing jitter within an iteration) is supplied, yet this directly scales the steady-state response amplitude used to obtain the safety bound.

Authors: The square-wave model follows directly from the deterministic Bulk Synchronous Parallel protocol described in the literature, with amplitude set from published GPU power traces. We did not perform a stochastic jitter analysis because the paper focuses on the periodic component that can drive resonance. We accept that real timing variations would alter the harmonic spectrum and could reduce peak amplitude at exact resonance. We will add a brief sensitivity study (new subsection or appendix) quantifying the effect of representative jitter levels on the steady-state response. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation inverts external two-area swing model with independent planning-study parameters and GPU specs

full rationale

The paper's central derivation inverts the steady-state forced response of the standard two-area swing equations, taking modal parameters directly from external planning-study eigenanalysis and forcing amplitude from GPU power specifications. These are independent inputs rather than quantities fitted or defined inside the paper. No self-citations, self-definitional steps, fitted-input predictions, or ansatz smuggling are present in the abstract or derivation description. The result (e.g., 66,900 GPU bound) is a direct algebraic consequence of those external parameters and the inversion, not a renaming or tautology. This is the most common honest non-finding for model-based screening criteria.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The claim rests on the standard two-area swing equations and the domain assumption of lockstep GPU behavior; modal parameters and forcing amplitude are drawn from external sources rather than derived or fitted within the paper.

free parameters (2)

modal response parameters
Taken from planning-study eigenanalysis
forcing amplitude
Taken from GPU specifications and duty cycle

axioms (2)

domain assumption Aggregate cluster power follows a square wave at the iteration period
Stated as every GPU transitions in lockstep under BSP
standard math Steady-state forced response of the two-area swing equations governs the deviation
Criterion obtained by inverting these equations

pith-pipeline@v0.9.1-grok · 5772 in / 1416 out tokens · 44952 ms · 2026-06-26T11:43:15.261140+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

23 extracted references · 3 canonical work pages

[1]

Kundur,Power System Stability and Control

P. Kundur,Power System Stability and Control. New York: McGraw- Hill, 1994

1994
[2]

Characteristics and risks of emerging large loads: Large loads task force white paper,

NERC, “Characteristics and risks of emerging large loads: Large loads task force white paper,” NERC, Tech. Rep., Jul. 2025

2025
[3]

A bridging model for parallel computation,

L. G. Valiant, “A bridging model for parallel computation,”Commun. ACM, vol. 33, no. 8, pp. 103–111, 1990

1990
[4]

Power stabilization for AI training datacenters,

E. Choukse, B. Warrier, S. Heath, L. Belmont, A. Zhao, H. A. Khan, B. Harry, M. Kappel, R. J. Hewett, K. Dattaet al., “Power stabilization for AI training datacenters,”arXiv preprint arXiv:2508.14318, 2025

arXiv 2025
[5]

Characterizing the efficiency of distributed training: A power, performance, and thermal perspective,

S. Go, J. Park, S. More, H. Wu, I. Wang, A. Jezghani, T. Krishna, and D. Mahajan, “Characterizing the efficiency of distributed training: A power, performance, and thermal perspective,” inProc. 58th IEEE/ACM Int. Symp. on Microarchitecture, 2025

2025
[6]

The unseen AI disruptions for power grids: LLM-induced transients,

Y . Li, M. Mughees, Y . Chen, and Y . R. Li, “The unseen AI disruptions for power grids: LLM-induced transients,”arXiv:2409.11416, 2024

arXiv 2024
[7]

Essential action to industry: Computational load modeling, studies, instrumentation, commissioning, operations, protection, and control,

NERC, “Essential action to industry: Computational load modeling, studies, instrumentation, commissioning, operations, protection, and control,” NERC, Level 3 NERC Alert, May 2026

2026
[8]

Industry recommendation: Large load interconnection, study, commissioning, and operations,

NERC, “Industry recommendation: Large load interconnection, study, commissioning, and operations,” North American Electric Reliability Corporation, Level 2 NERC Alert, Sep. 2025

2025
[9]

Wide-area power system oscillations from large- scale AI workloads,

M.-S. Ko and H. Zhu, “Wide-area power system oscillations from large- scale AI workloads,”IEEE Transactions on Power Systems, pp. 1–14, 2026

2026
[10]

Operational risks in grid integration of large data center loads: Characteristics, stability assessments, and sensitivity studies,

K.-B. Kwon, S. Mukherjee, and V . Adetola, “Operational risks in grid integration of large data center loads: Characteristics, stability assessments, and sensitivity studies,”arXiv:2510.05437, 2025

arXiv 2025
[11]

Spatial load correlation in AI data-center-dominated power systems,

C. Chaudhary, A. Abdelkader, Y . Pei, M. Ben-Idris, and J. Mitra, “Spatial load correlation in AI data-center-dominated power systems,” in Proc. 2026 IEEE Power & Energy Society General Meeting (PES GM), Montr´eal, QC, Canada, Jul. 2026, preprint: https://doi.org/10.13140/RG. 2.2.28516.13442

work page doi:10.13140/rg 2026
[12]

Modal analysis of spatial load correlation in AI data center- dominated power systems,

C. Chaudhary, M. Murillo, M. Ben-Idris, J. Mitra, D. Pandit, and A. Bera, “Modal analysis of spatial load correlation in AI data center- dominated power systems,” inProc. IEEE Int. Conf. on Smart Energy Systems and Technologies (SEST), September 2026, preprint: https: //doi.org/10.13140/RG.2.2.17610.94404

work page doi:10.13140/rg.2.2.17610.94404 2026
[13]

The foundations of locale theory

C. Chaudhary, A. Abdelkader, M. Ben-Idris, and J. Mitra, “Resource adequacy risk in correlated large loads,” inProc. IEEE Int. Conf. on Probabilistic Methods Applied to Power Systems (PMAPS), Salt Lake City, UT, USA, Sep. 2026, preprint: https://doi.org/10.13140/RG.2.2. 35227.02087

work page doi:10.13140/rg.2.2 2026
[14]

Data center power equipment thermal guidelines and best practices,

ASHRAE Technical Committee 9.9, “Data center power equipment thermal guidelines and best practices,” ASHRAE, Tech. Rep., 2016

2016
[15]

Rogers,Power System Oscillations

G. Rogers,Power System Oscillations. Boston, MA: Kluwer Academic Publishers, 2000

2000
[16]

Impact of data center load modeling on power system stability,

C. Chaudhary, A. Abdelkader, M. Egan, E. Udren, M. Ben-Idris, and J. Mitra, “Impact of data center load modeling on power system stability,” inGrid of the Future Symposium, ser. CIGRE US, Denver, Colorado, USA, Nov. 2025

2025
[17]

Understanding the inception of 14.7 Hz oscillations emerging from a data center,

C. Mishra, L. Vanfretti, J. Delaree Jr., T. J. Purcell, and K. D. Jones, “Understanding the inception of 14.7 Hz oscillations emerging from a data center,”Sustainable Energy, Grids and Networks, vol. 43, p. 101735, 2025

2025
[18]

A practical method for the direct analysis of transient stability,

T. Athay, R. Podmore, and S. Virmani, “A practical method for the direct analysis of transient stability,”IEEE Trans. Power App. Syst., vol. PAS-98, no. 2, pp. 573–584, 1979

1979
[19]

M. A. Pai,Energy Function Analysis for Power System Stability. Boston, MA: Kluwer Academic Publishers, 1989

1989
[20]

Bench- mark models for the analysis and control of small-signal oscillatory dynamics in power systems,

C. Canizares, T. Fernandes, E. Geraldi, L. Gerin-Lajoie, M. Gibbard, I. Hiskens, J. Kersulis, R. Kuiava, L. Lima, F. DeMarcoet al., “Bench- mark models for the analysis and control of small-signal oscillatory dynamics in power systems,”IEEE Transactions on Power Systems, vol. 32, no. 1, pp. 715–722, 2016

2016
[21]

Interconnection oscillation analysis: Reliability assessment,

NERC, “Interconnection oscillation analysis: Reliability assessment,” North American Electric Reliability Corporation, Tech. Rep., 2019

2019
[22]

Standard BAL-001-2: Real power balancing control perfor- mance,

NERC, “Standard BAL-001-2: Real power balancing control perfor- mance,” NERC, NERC Reliability Standard, 2015

2015
[23]

Measurement adequacy for monitoring data center oscillations,

K. Chatterjee, J. D. Follum, A. Varghese, S. Biswas, E. Farantatos, and L. Zhu, “Measurement adequacy for monitoring data center oscillations,” Pacific Northwest National Laboratory, Tech. Rep., 2026

2026

[1] [1]

Kundur,Power System Stability and Control

P. Kundur,Power System Stability and Control. New York: McGraw- Hill, 1994

1994

[2] [2]

Characteristics and risks of emerging large loads: Large loads task force white paper,

NERC, “Characteristics and risks of emerging large loads: Large loads task force white paper,” NERC, Tech. Rep., Jul. 2025

2025

[3] [3]

A bridging model for parallel computation,

L. G. Valiant, “A bridging model for parallel computation,”Commun. ACM, vol. 33, no. 8, pp. 103–111, 1990

1990

[4] [4]

Power stabilization for AI training datacenters,

E. Choukse, B. Warrier, S. Heath, L. Belmont, A. Zhao, H. A. Khan, B. Harry, M. Kappel, R. J. Hewett, K. Dattaet al., “Power stabilization for AI training datacenters,”arXiv preprint arXiv:2508.14318, 2025

arXiv 2025

[5] [5]

Characterizing the efficiency of distributed training: A power, performance, and thermal perspective,

S. Go, J. Park, S. More, H. Wu, I. Wang, A. Jezghani, T. Krishna, and D. Mahajan, “Characterizing the efficiency of distributed training: A power, performance, and thermal perspective,” inProc. 58th IEEE/ACM Int. Symp. on Microarchitecture, 2025

2025

[6] [6]

The unseen AI disruptions for power grids: LLM-induced transients,

Y . Li, M. Mughees, Y . Chen, and Y . R. Li, “The unseen AI disruptions for power grids: LLM-induced transients,”arXiv:2409.11416, 2024

arXiv 2024

[7] [7]

Essential action to industry: Computational load modeling, studies, instrumentation, commissioning, operations, protection, and control,

NERC, “Essential action to industry: Computational load modeling, studies, instrumentation, commissioning, operations, protection, and control,” NERC, Level 3 NERC Alert, May 2026

2026

[8] [8]

Industry recommendation: Large load interconnection, study, commissioning, and operations,

NERC, “Industry recommendation: Large load interconnection, study, commissioning, and operations,” North American Electric Reliability Corporation, Level 2 NERC Alert, Sep. 2025

2025

[9] [9]

Wide-area power system oscillations from large- scale AI workloads,

M.-S. Ko and H. Zhu, “Wide-area power system oscillations from large- scale AI workloads,”IEEE Transactions on Power Systems, pp. 1–14, 2026

2026

[10] [10]

Operational risks in grid integration of large data center loads: Characteristics, stability assessments, and sensitivity studies,

K.-B. Kwon, S. Mukherjee, and V . Adetola, “Operational risks in grid integration of large data center loads: Characteristics, stability assessments, and sensitivity studies,”arXiv:2510.05437, 2025

arXiv 2025

[11] [11]

Spatial load correlation in AI data-center-dominated power systems,

C. Chaudhary, A. Abdelkader, Y . Pei, M. Ben-Idris, and J. Mitra, “Spatial load correlation in AI data-center-dominated power systems,” in Proc. 2026 IEEE Power & Energy Society General Meeting (PES GM), Montr´eal, QC, Canada, Jul. 2026, preprint: https://doi.org/10.13140/RG. 2.2.28516.13442

work page doi:10.13140/rg 2026

[12] [12]

Modal analysis of spatial load correlation in AI data center- dominated power systems,

C. Chaudhary, M. Murillo, M. Ben-Idris, J. Mitra, D. Pandit, and A. Bera, “Modal analysis of spatial load correlation in AI data center- dominated power systems,” inProc. IEEE Int. Conf. on Smart Energy Systems and Technologies (SEST), September 2026, preprint: https: //doi.org/10.13140/RG.2.2.17610.94404

work page doi:10.13140/rg.2.2.17610.94404 2026

[13] [13]

The foundations of locale theory

C. Chaudhary, A. Abdelkader, M. Ben-Idris, and J. Mitra, “Resource adequacy risk in correlated large loads,” inProc. IEEE Int. Conf. on Probabilistic Methods Applied to Power Systems (PMAPS), Salt Lake City, UT, USA, Sep. 2026, preprint: https://doi.org/10.13140/RG.2.2. 35227.02087

work page doi:10.13140/rg.2.2 2026

[14] [14]

Data center power equipment thermal guidelines and best practices,

ASHRAE Technical Committee 9.9, “Data center power equipment thermal guidelines and best practices,” ASHRAE, Tech. Rep., 2016

2016

[15] [15]

Rogers,Power System Oscillations

G. Rogers,Power System Oscillations. Boston, MA: Kluwer Academic Publishers, 2000

2000

[16] [16]

Impact of data center load modeling on power system stability,

C. Chaudhary, A. Abdelkader, M. Egan, E. Udren, M. Ben-Idris, and J. Mitra, “Impact of data center load modeling on power system stability,” inGrid of the Future Symposium, ser. CIGRE US, Denver, Colorado, USA, Nov. 2025

2025

[17] [17]

Understanding the inception of 14.7 Hz oscillations emerging from a data center,

C. Mishra, L. Vanfretti, J. Delaree Jr., T. J. Purcell, and K. D. Jones, “Understanding the inception of 14.7 Hz oscillations emerging from a data center,”Sustainable Energy, Grids and Networks, vol. 43, p. 101735, 2025

2025

[18] [18]

A practical method for the direct analysis of transient stability,

T. Athay, R. Podmore, and S. Virmani, “A practical method for the direct analysis of transient stability,”IEEE Trans. Power App. Syst., vol. PAS-98, no. 2, pp. 573–584, 1979

1979

[19] [19]

M. A. Pai,Energy Function Analysis for Power System Stability. Boston, MA: Kluwer Academic Publishers, 1989

1989

[20] [20]

Bench- mark models for the analysis and control of small-signal oscillatory dynamics in power systems,

C. Canizares, T. Fernandes, E. Geraldi, L. Gerin-Lajoie, M. Gibbard, I. Hiskens, J. Kersulis, R. Kuiava, L. Lima, F. DeMarcoet al., “Bench- mark models for the analysis and control of small-signal oscillatory dynamics in power systems,”IEEE Transactions on Power Systems, vol. 32, no. 1, pp. 715–722, 2016

2016

[21] [21]

Interconnection oscillation analysis: Reliability assessment,

NERC, “Interconnection oscillation analysis: Reliability assessment,” North American Electric Reliability Corporation, Tech. Rep., 2019

2019

[22] [22]

Standard BAL-001-2: Real power balancing control perfor- mance,

NERC, “Standard BAL-001-2: Real power balancing control perfor- mance,” NERC, NERC Reliability Standard, 2015

2015

[23] [23]

Measurement adequacy for monitoring data center oscillations,

K. Chatterjee, J. D. Follum, A. Varghese, S. Biswas, E. Farantatos, and L. Zhu, “Measurement adequacy for monitoring data center oscillations,” Pacific Northwest National Laboratory, Tech. Rep., 2026

2026