Pith. sign in

REVIEW 3 major objections 4 minor 19 references

A data center's UPS channel can act as a fast demand-response resource that adds virtual damping to inter-area oscillations, lifting the critical mode's damping ratio by 73.7% at an optimized gain, while the HVAC subsystem is too slow to he

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 05:21 UTC pith:AW6OXQSH

load-bearing objection A credible qualitative case that UPS fast demand response can damp inter-area modes while HVAC cannot, but the quantitative 73.7% improvement is not validated because the actuator saturates in the time-domain test and one validation point is physically impossible. the 3 major comments →

arxiv 2607.27575 v1 pith:AW6OXQSH submitted 2026-07-30 eess.SY cs.SY

Inter-Area Oscillation Damping in Data-Center-Integrated Power Systems

Ahmed Alfatlawi , Masoud H. Nazari This is my paper
classification eess.SY cs.SY
keywords data center demand responseinter-area oscillationssmall-signal stabilityvirtual dampingUPS fast channelHVAC thermal dynamicseigenvalue sensitivityIEEE 39-bus system
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Hyperscale data centers are becoming transmission-scale loads, and this paper asks whether their internal equipment can help damp inter-area oscillations. Treating the data center not as a single aggregated load but as three distinct timescales, the authors show analytically that only the UPS subsystem has the bandwidth to track 1–2 Hz grid oscillations. A proportional frequency-feedback controller on the UPS adds a virtual damping term to the generator swing equation, and a gradient-based eigenvalue-sensitivity optimization selects the gain that maximizes the critical mode's damping ratio, improving it from 1.29% to about 2.24% in the IEEE 39-bus system. The HVAC cooling plant, with an 8-second thermal time constant, attenuates the 1.94 Hz signal by roughly 99% and is declared physically incapable of this task. A sympathetic reader would take the paper's main claim as: choose demand-response resources by matching their time constants to the oscillation band being targeted.

Core claim

On its own terms, the paper's central claim is that UPS-based demand response can enhance inter-area oscillation damping, whereas HVAC-based demand response cannot, and the reason is bandwidth, not tuning. The UPS injection creates additive virtual damping D_DR equal to the feedback gain, because power is modulated in tens of milliseconds; the HVAC's first-order thermal dynamics act as a low-pass filter with magnitude 0.0103 at 1.94 Hz. With a 100 MW step disturbance on the IEEE 39-bus system, the UPS channel saturates at its 5.6 MW hardware limit while the HVAC channel delivers zero response; the settling time improves from 14.39 s to 10.48 s and the damping ratio of the critical inter-area

What carries the argument

Additive virtual damping from a frequency-proportional demand-response channel: substituting ∆P_DR = -K_DR,ω Δω_COI into the aggregated swing equation yields an effective damping coefficient D + D_DR with D_DR = K_DR,ω. The paper embeds this in a 47-state augmented small-signal model, models the UPS as a 50 ms first-order actuator with a lead-lag compensator zeroing its phase lag at 1.94 Hz, models the HVAC as a coupled two-state thermal system with an 8 s time constant, and tunes K_UPS via a gradient-based eigenvalue-sensitivity algorithm.

Load-bearing premise

The 73.7% damping improvement is computed with an unsaturated UPS controller, but the time-domain validation saturates the UPS at its 5.6 MW hardware limit, so the additive virtual damping D_DR=K does not describe the operating regime used to verify the transient response.

What would settle it

Run the same 100 MW step disturbance and eigenvalue analysis with a saturation-aware UPS model (e.g., a limited integrator or a describing-function approximation of the 5.6 MW limit). If the saturated model's critical-mode damping ratio or settling time differs substantially from the reported 2.24% and 10.48 s, the quantified benefit is an artifact of the unsaturated assumption.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • A firmware-level gain adjustment on an existing UPS can provide inter-area damping without additional transmission hardware.
  • HVAC-based demand response should not be counted on for inter-area oscillation damping; its contribution is negligible at 0.1–2 Hz.
  • Data centers can be modeled for stability studies with distinct subsystem timescales rather than a single aggregated load.
  • The optimized gain raises the critical inter-area damping ratio from 1.29% to roughly 2.24%, a 73.7% improvement.
  • Without the lead-lag compensation, the 50 ms UPS actuator would lag by about 31 degrees at 1.94 Hz and reduce the damping contribution.

Where Pith is reading between the lines

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

  • [editorial inference] Because the time-domain test saturates the UPS at 5.6 MW while the damping-ratio calculation assumes an unsaturated proportional law, the 73.7% figure should be re-checked with a saturating nonlinearity; the qualitative benefit likely survives, but the exact number may not.
  • [editorial inference] The bandwidth argument generalizes as a selection rule: any fast power-electronics load (EV chargers, battery storage, some industrial drives) could provide similar virtual damping, while slower thermal loads cannot—this is a testable extension.
  • [editorial inference] The paper leaves IT workload throttling as future work; since the UPS headroom is preserved at 80% for ride-through, IT throttling is the natural complementary fast channel and may extend the damping range.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. Alfatlawi and Nazari develop a 47-state small-signal model of the IEEE 39-bus system with an explicit HVAC and UPS model for a 700 MW data center. They show analytically that the HVAC channel attenuates a 1.94 Hz inter-area signal by 98.97% (|G|=0.0103) and design a COI-frequency-proportional UPS demand-response controller with lead-lag compensation. A gradient-based optimization selects K_UPS=3571, raising the critical mode damping ratio from 1.29% to 2.24% (a 73.7% improvement). Time-domain 100 MW step simulations are presented as validation, reporting reduced settling time, peak COI deviation, and modal energy, with the UPS channel saturated at its 5.6 MW limit throughout.

Significance. The paper addresses a timely application—data centers as providers of inter-area oscillation damping—and the explicit two-subsystem decomposition is a useful modeling contribution. The HVAC bandwidth attenuation calculation is transparent and its conclusion (the slow thermal loop cannot track 1.94 Hz oscillations) is robust. The UPS fast-channel mechanism is physically plausible, and the small-signal framework is standard. However, the quantitative headline (73.7% damping improvement) rests on an eigenvalue calculation for an unsaturated controller, while the time-domain validation reports immediate saturation. The transient benefits are therefore not shown to be caused by the additive virtual damping D_DR=K; they are produced by a saturating nonlinear control signal. This is a central validation gap requiring major revision. If the unsaturated linear prediction and a saturation-aware transient analysis are reconciled, the contribution would be significant for data-center-integrated grid stability studies.

major comments (3)
  1. [Sec. III-C / Sec. IV] The paper claims a 73.7% damping-ratio improvement (ζ: 1.29%→2.24%) from the eigenvalue analysis at K_UPS=3571, and presents the time-domain results in Table III as validation. However, Table III states that the UPS channel 'saturates immediately at its 5.6 MW hardware limit.' With ΔP_DR=-KΔω_COI, saturation begins at |Δω|=5.6MW/3571=1.57×10^-5 pu≈0.00094 Hz. The 100 MW step produces peak COI deviations of ~0.19 Hz (Table III), several orders of magnitude larger; the transient never operates in the proportional regime where D_DR=K applies. The settling-time and modal-energy improvements in Table III are consequences of a saturating (effectively bang-bang) injection, not of the additive virtual damping derived in Section III-C. Please validate with a gain small enough to remain unsaturated for the tested disturbance, or analyze the saturated closed loop explicitly and separate the two mec
  2. [Sec. II-A / Table II / Eq. (22)] There are conflicting statements about the UPS demand-response range. Section II-A and Table I state a ±5.6 MW range (20% of 28 MW), but Eq. (22) imposes |ΔP_UPS|≤0.8P_UPS (22.4 MW). Table II then lists a 'near-equivalent validation point (K=375, P=50MW)' with P=50 MW, which exceeds the stated hardware limit by an order of magnitude. If P in Table II is not the injection range, define it; if it is, the near-equivalent point is infeasible. Since the feasibility of K_UPS=3571 under the 5.6 MW limit is central to the saturation argument, this inconsistency must be resolved.
  3. [Sec. III-C, Eqs. (17)–(18)] The 'formal derivation' of virtual damping is not a derivation: substituting the control law ΔP_DR=-KΔω_COI into (16) and collecting terms defines D_DR=K by construction. The statement 'This result implies that the contribution depends only on the controller gain' is tautological. What needs justification is whether the UPS actuator plus compensator preserves the sign of the real part at 1.94 Hz (Eq. (24))—that part is sound—and whether the control is unsaturated. The current presentation overstates the contribution of Eqs. (17)–(18); please rephrase as a modeling convention and let the eigenvalue sensitivity carry the physical claim.
minor comments (4)
  1. [Eq. (15)] The block matrix repeats 'A_g,dc and A_g,dc'; one of these should presumably be A_dc,g.
  2. [Sec. III-B / Eq. (14)] The three 'slow controller states' Δx_ctrl in R^3 are never defined. Please specify their differential equations (e.g., lead-lag compensator states and any measurement filter).
  3. [Algorithm 1 / Eq. (25)] Eq. (25) defines dJ/dK, but Algorithm 1 labels g as dζ_min/dK and says 'gradient ascent.' Since J=-ζ_min, minimizing J is equivalent to maximizing ζ_min; the sign/wording should be corrected for consistency.
  4. [References] Reference [18] duplicates [10]; 'overll' typo in Section V.

Circularity Check

3 steps flagged

The damping improvement is partly by construction: D_DR is the control-law gain, and the time-domain 'confirmations' rely on UPS saturation and a disabled HVAC channel.

specific steps
  1. self definitional [Section III-C, Eqs. (16)-(18)]
    "The demand-response controller generates a signal proportional to the COI frequency deviation, denoted Δω_COI: ΔP_DR = −K_DR,ω Δω_COI ≈ −K_DR,ω Δω ... Substituting into (16) yields: MΔω̇ = ... −(D + D_DR)Δω ... The demand-response contribution appears as an additive virtual damping coefficient D_DR = K_DR,ω."

    D_DR is not derived from independent electromechanical physics; it is the gain in the control law by definition. Substituting a proportional feedback law into the swing equation is an algebraic identity, so the 'additive virtual damping' is an input assumption of the design, not a first-principles finding. The paper then presents the resulting eigenvalue improvement as a demonstration of UPS effectiveness, which is the same feedback inserted into the model.

  2. other [Section IV, Table III and Fig. 6]
    "The UPS channel saturates immediately at its 5.6 MW hardware limit, confirming full actuator utilization and that the virtual damping coefficient D_DR operates at its maximum."

    The linear eigenvalue improvement and D_DR=K assume ΔP_DR = −KΔω, but the time-domain simulation's UPS output is a constant ±5.6 MW (saturated). A saturated relay is not proportional to frequency, so the observed transient cannot confirm the linear virtual-damping mechanism; it confirms only the imposed saturation limit. The validation thus uses a regime in which the claimed damping coefficient is absent.

  3. other [Section IV, Table III and Section II scenario definitions]
    "The HV AC channel contributes zero demand response throughout the entire transient, confirming the bandwidth attenuation demonstrated analytically in Section III-C."

    Scenario 3 only activates UPS-based fast-channel demand response; no HVAC demand-response command is applied. Hence zero HVAC output is built into the scenario, not an independent confirmation of the thermal bandwidth limit. The analytical 0.0103 magnitude at 1.94 Hz is valid, but the time-domain 'confirmation' is by construction.

full rationale

The paper builds an explicit 47-state small-signal model and uses a standard IEEE 39-bus benchmark, so the core modeling is self-contained and does not rest on a self-citation chain. The main circularity concern is that the quantitative damping claim reduces to the controller design: the 'virtual damping coefficient' D_DR = K is obtained by substituting the proportional control law ΔP_DR = −KΔω into the swing equation, making the damping contribution an algebraic identity rather than an independent result. The 73.7% improvement is then the objective of the gradient-based optimization, so reporting it as a demonstrated benefit is partly a restatement of the optimization goal. The time-domain validation is compromised: the UPS saturates immediately, so the simulated transient does not actually exercise the proportional law used to compute the eigenvalue improvement, and the HVAC channel is never commanded, so its zero output is by construction. These are partial circularities in the validation chain, not complete ones, because the qualitative conclusion that a fast UPS channel can contribute more damping than a slow thermal channel is supported by the bandwidth calculation and by plausible linear analysis. Score 5 reflects partial circularity: the central numerical demonstration is substantially by construction, while the modeling and qualitative physical argument retain independent content.

Axiom & Free-Parameter Ledger

7 free parameters · 4 axioms · 0 invented entities

The central quantitative claims rest on several hand-set parameters, most consequentially K_UPS (optimized), T_HVAC, and D. No new physical entities are introduced; the "virtual damping" is a relabeling of the frequency-proportional control law.

free parameters (7)
  • UPS controller gain K_UPS = 3571 (per-unit on 100 MVA base)
    Chosen by gradient optimization (Eqs. 19-25) to maximize ζ_min; the damping-ratio claims depend on it.
  • HVAC thermal time constant T_HVAC = 8.0 s
    Assumed from ASHRAE/Patterson; drives the 98.97% attenuation and the "inherently incapable" conclusion.
  • Generator damping D = 2 pu
    Set uniformly across all 10 machines; the baseline ζ=1.29% is a function of this choice.
  • UPS actuator and compensator time constants τ_UPS, T_lead, T_lag = 0.05 s, 0.085 s, 0.020 s
    Chosen so Re(G_comp)=1.194 at 1.94 Hz; the claimed achievable virtual damping is sensitive to these values.
  • Data-center load decomposition = IT 60%, HVAC 30%, aux 6%, UPS 4% of 700 MW
    Assumed from PUE literature; fixes the DR capacities (5.6 MW UPS, 42 MW HVAC).
  • UPS DR range as 20% of UPS capacity = ±5.6 MW
    Assumed headroom for ride-through; contradicts constraint (22), which as written allows 0.8*P_UPS = 22.4 MW.
  • HVAC thermal model coefficients C_th, R_th, K_th, γ = 5.0, 1.0, 1.8, 2.0
    Hand-set values for the two-state thermal model (Eqs. 11-12); no data fitting or sensitivity study is provided.
axioms (4)
  • standard math The classical swing equation, governor, exciter, and network linearization (Eqs. 2-7) provide a valid small-signal model of the 10-machine IEEE 39-bus system.
    Standard power system modeling; the paper relies on it for all eigenvalue results.
  • domain assumption Data-center subsystems are representable by first-order linear ODEs with the stated time constants (Eqs. 10-13).
    The HVAC and UPS models are simplified representations, not validated against measured hardware data.
  • domain assumption A wide-area COI frequency signal is available and can be fed back to UPS power electronics with negligible communication delay.
    The controller (Eq. 17) uses Δω_COI without modeling latency, packet drop, or measurement noise; these would alter the phase at 1.94 Hz.
  • domain assumption A frequency-proportional active-power injection acts as additive damping in the aggregated swing equation (Eqs. 16-18).
    This is the control-law definition plus linearization; the paper treats it as a derived result rather than an assumption.

pith-pipeline@v1.3.0-daily-deepseek · 7119 in / 15548 out tokens · 143980 ms · 2026-08-01T05:21:55.788740+00:00 · methodology

0 comments
read the original abstract

This paper develops explicit dynamic models of a hyperscale data center, including its heating, ventilation, and air conditioning (HVAC) and uninterruptible power supply (UPS) subsystems, and integrates them into a small-signal stability framework to investigate the impact of data center demand response on power system inter-area oscillations. Through eigenvalue analysis and time-domain simulations, the results demonstrate that UPS-based demand response can enhance inter-area oscillation damping. In contrast, the HVAC subsystem is shown to be inherently incapable of providing effective oscillation damping due to its limited thermal response bandwidth. A gradient-based optimization algorithm is used to tune the UPS controller gain to maximize the damping ratio of the critical inter-area mode. The effectiveness of the proposed approach is validated using the IEEE 39-bus test system.

Figures

Figures reproduced from arXiv: 2607.27575 by Ahmed Alfatlawi, Masoud H. Nazari.

Figure 2
Figure 2. Figure 2: Steady-state bus voltage profiles (kV) for all three scenarios, with [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: IEEE 39-bus New England test system with 700 MW data center split [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sensitivity of ζmin to UPS gain KUPS at P = 5.6 MW. TABLE III TIME-DOMAIN PERFORMANCE METRICS (100 MW STEP DISTURBANCE) Metric S2 (no DR) S3 (DR) Impr. Settling time (s) 14.39 10.48 −27.2% Peak COI frequency deviation (Hz) 0.2233 0.1929 −13.6% Frequency nadir (Hz) 59.777 59.807 +0.030 Modal energy 3.678×10−3 1.808×10−3 −50.9% Peak UPS DR (MW) — 5.600 100% sat. Peak HVAC DR (MW) — 0.000 0% [PITH_FULL_IMAGE… view at source ↗
Figure 3
Figure 3. Figure 3: Minimum inter-area damping ratio across the three studied scenarios. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 6
Figure 6. Figure 6: UPS demand-response power output during Scenario 3. [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The COI frequency response following the 100 MW load-step [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Settling-time comparison using the 2% criterion. [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

19 extracted references · 1 linked inside Pith

  1. [1]

    Recalibrating global data center energy-use estimates,

    E. Masanet, A. Shehabi, N. Lei, S. Smith, and J. Koomey, “Recalibrating global data center energy-use estimates,”Science, vol. 367, no. 6481, pp. 984–986, Feb. 2020

    2020

  2. [2]

    Data Centres and Data Transmission Networks,

    International Energy Agency, “Data Centres and Data Transmission Networks,” IEA, Paris, Tech. Rep., 2024. [Online]. Available: https: //www.iea.org/reports/data-centres-and-data-transmission-networks

    2024

  3. [3]

    Data center power system stability—Part I: Power supply impedance modeling,

    J. Sun, M. Xu, M. Cespedes, and M. Kauffman, “Data center power system stability—Part I: Power supply impedance modeling,”CSEE J. Power Energy Syst., vol. 8, no. 2, pp. 403–419, Mar. 2022

    2022

  4. [4]

    Data center power system stability—Part II: System modeling and analysis,

    J. Sun, M. Xu, M. Cespedes, and M. Kauffman, “Data center power system stability—Part II: System modeling and analysis,”CSEE J. Power Energy Syst., vol. 8, no. 2, pp. 420–431, Mar. 2022

    2022

  5. [5]

    Kundur,Power System Stability and Control

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

    1994

  6. [6]

    Rogers,Power System Oscillations

    G. Rogers,Power System Oscillations. Norwell, MA, USA: Kluwer Academic, 2000

    2000

  7. [7]

    Learning-based distributed optimal power sharing and frequency control under cyber contingencies,

    S. Xie, M. H. Nazari, and L. Y . Wang, “Learning-based distributed optimal power sharing and frequency control under cyber contingencies,” Int. J. Electr . Power Energy Syst., vol. 152, 2023

    2023

  8. [8]

    Machowski, Z

    J. Machowski, Z. Lubosny, J. W. Bialek, and J. R. Bumby,Power System Dynamics: Stability and Control, 3rd ed. Hoboken, NJ, USA: Wiley, 2020

    2020

  9. [9]

    P. W. Sauer, M. A. Pai, and J. H. Chow,Power System Dynamics and Stability, 2nd ed. Hoboken, NJ, USA: Wiley-IEEE Press, 2017

    2017

  10. [10]

    M. H. Nazari, M. Ilic, J. P. Lopes, Small-signal stability and decentral- ized control design for electric energy systems with a large penetration of distributed generators, Control Engineering Practice, V olume 20, Issue 9, 2012, Pages 823-831

    2012

  11. [11]

    Pricing data center demand response,

    Z. Liu, I. Liu, S. Low, and A. Wierman, “Pricing data center demand response,” inProc. ACM SIGMETRICS, Pittsburgh, PA, USA, 2013, pp. 111–122

    2013

  12. [12]

    Decentralized demand-side contribution to primary frequency control,

    J. D. Molina-Garcia, F. Bouffard, and D. S. Kirschen, “Decentralized demand-side contribution to primary frequency control,”IEEE Trans. Power Syst., vol. 26, no. 1, pp. 411–419, Feb. 2011

    2011

  13. [13]

    Data centres as a source of flexibility for power systems,

    M. T. Takci, M. Qadrdan, J. Summers, and J. Gustafsson, “Data centres as a source of flexibility for power systems,”Energy Rep., vol. 13, pp. 3661–3671, 2025

    2025

  14. [14]

    Dynamic modeling of a data center for power system stability studies,

    P. P. Gyang, P. Chakraborty, L. Meegahapola, and X. Yu, “Dynamic modeling of a data center for power system stability studies,”IEEE Trans. Power Syst., 2025

    2025

  15. [15]

    Milano,Power System Modelling and Scripting

    F. Milano,Power System Modelling and Scripting. London, UK: Springer, 2010

    2010

  16. [16]

    Atlanta, GA, USA: ASHRAE, 2021

    ASHRAE Technical Committee 9.9,Data Center Power Equipment Thermal Guidelines and Best Practices. Atlanta, GA, USA: ASHRAE, 2021

    2021

  17. [17]

    The effect of data center temperature on energy efficiency,

    M. K. Patterson, “The effect of data center temperature on energy efficiency,” inProc. 11th Intersoc. Conf. Therm. Thermomech. Phenom. Electron. Syst., Orlando, FL, USA, 2008, pp. 1167–1174

    2008

  18. [18]

    Small-signal stability and decentral- ized control design for electric energy systems with a large penetration of distributed generators,

    M. H. Nazari, M. Ilic, J. P. Lopes, “Small-signal stability and decentral- ized control design for electric energy systems with a large penetration of distributed generators,”Control Engineering Practice, V olume 20, Issue 9, 2012, Pages 823-831

    2012

  19. [19]

    Coordinating Power Grid Frequency Regulation Service with Data Center Load Flexibility,

    A. Jahanshahi, S. R. Golrouye, O. Anderson, N. Yu, and D. Wong, “Coordinating Power Grid Frequency Regulation Service with Data Center Load Flexibility,”arXiv preprint arXiv:2601.22487, Jan. 2026

    Pith/arXiv arXiv 2026