Skip to content
TheChatArchiveThe Chat Archive
o3-2025-04-16· OpenAIHugging Face LMArena Human Preference 140K

as a geophysicist and DSP engineer, tell in dispersion curve plot - velocity dependency…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Googlegemma-3-27b-it’s answer.
  1. User

    as a geophysicist and DSP engineer, tell in dispersion curve plot - velocity dependency of frequency. What type of velocity is meant - phase or group?

  2. Assistant

    When you see a “dispersion curve” in near-surface seismology / MASW / SASW papers (or in ocean acoustics, optics, etc.), the default assumption is almost always:

    Phase velocity (cₚ = ω / k) plotted as a function of frequency (or its reciprocal, period).

    Why phase and not group?

    1. How the data are extracted
      • F-k, MASW, ReMi, spatial‐autocorrelation, wave‐number beamforming, etc. all pick the slope of the spatial phase term e^{j(kx – ωt)}. The slope in the x–t domain gives k, so you get cₚ = ω / k directly.
      • Multiple-filter or frequency–time (FTAN) methods do recover group velocity first, but they are usually converted to phase velocity before inversion, because earth models are more directly related to cₚ(ω).

    2. What inversion codes expect
      In surface-wave inversion (Herrmann’s Computer Programs in Seismology, Dinver, SurfTomo, etc.) the forward solver computes the phase-velocity dispersion of Rayleigh/Love modes. If someone handed you a group-velocity curve, you would have to convert it (using cₚ and modal derivatives) before feeding it to the inversion.

    3. Convention in figures
      If a paper simply writes “dispersion curve” and shows velocity vs. frequency/period/wavelength, 99 % of the time it’s phase velocity. When the author means group velocity they label the axis explicitly (c_g) or write “group dispersion” because they know the convention is phase.

    Quick refresher:

    Phase velocity cₚ(ω) = ω / k(ω)
    Group velocity c_g(ω) = dω / dk = cₚ – ω (∂cₚ/∂ω)

    In a non-dispersive medium cₚ = c_g, so the distinction vanishes, but in the real earth Rayleigh and Love waves are strongly dispersive and cₚ ≠ c_g.

    Edge cases

    • Wideband FTAN plots: the first panel is group velocity vs period; a second panel or over-plotted curve may show the converted phase velocity.
    • Long-period, global tomography: Both cₚ(T) and c_g(T) maps are published, with labels.
    • EM or optical dispersion diagrams often show both β (propagation constant) and v_g; again they are clearly labeled.

    Take-away

    Unless the figure axis or caption says “group velocity”, interpret a dispersion curve as phase velocity. If you are preparing a plot yourself, add a subscript (cₚ or c_g) so readers don’t have to guess.

  3. User

    What should i expect for V_R (Rayleigh waves velocity)?

  4. Assistant

    Rayleigh-wave velocity, V_R, is not a fixed number—it depends on the elastic parameters of the medium and on frequency (layering ⇒ dispersion).
    Still, there are a few “anchor points” and rules of thumb that let you sanity-check any dispersion curve or inversion result:

    ────────────────────────────────────────────

    1. Homogeneous half-space (single layer)

    For an isotropic, elastic half-space with P- and S-wave velocities V_P and V_S:

    a. Exact (secular) equation
    Let
    β = (V_S / V_P)² = 1 − 2ν (ν = Poisson’s ratio)
    ζ = V_R / V_S
    Then
    ζ⁶ − 8ζ⁴ + 8(3 − 2β)ζ² − 16(1 − β) = 0

    b. Handy approximations
    • Ewing et al. (1957) numerical fit:
    V_R ≈ V_S · (0.862 + 1.14β) (|error| < 0.1 %)
    • AVO textbooks sometimes quote
    V_R ≈ 0.92 V_S (valid for ν ≈ 0.25, the “quasi-Poisson” earth)

    c. Ratio V_R / V_S vs Poisson’s ratio

    ν (Poisson)β = (V_S/V_P)²V_R / V_S
    01.000.955
    0.100.830.939
    0.200.670.925
    0.25 (common)0.600.919
    0.300.560.912
    0.330.500.905
    0.400.330.890
    0.450.220.860

    Take-away: in a homogeneous medium V_R is always a bit slower than V_S, typically 0.87–0.96 × V_S.

    ──────────────────────────────────────────── 2. Layered earth → dispersion

    Near-surface MASW / SASW / ReMi work deals with strong velocity gradients:

    • At high frequencies (short wavelengths) Rayleigh waves “feel” only the shallow, slow layers ⇒ V_R can be as low as 80–150 m/s in soft soils or pavement sub-grades.
    • At lower frequencies (long wavelengths) they sample deeper, stiffer material ⇒ V_R climbs toward the half-space V_S, reaching several hundred m/s in soil sites and > 3 km/s in crystalline bedrock.

    A typical fundamental-mode dispersion curve might look like this:

    Frequency (Hz) 20 10 5 2 1 0.5
    λ ≈ V_R/f (m) 30 60 120 300 600 1200
    V_R (m/s) 150 200 300 450 550 650

    (Exact numbers depend on the site’s layering, but the upward trend—Rayleigh waves speed up with decreasing frequency—is the hallmark of a low-velocity surface layer over faster basement.)

    ──────────────────────────────────────────── 3. Practical checkpoints

    1. V_R can never exceed V_P nor V_S of the half-space.
    2. For soft soil, if you invert a dispersion curve and get V_R > 0.96 V_S anywhere, something is wrong—usually mode mis-picking or a units mix-up.
    3. If your low-frequency V_R flattens out, that plateau ≈ the basement V_S (times the ratio in the table).
    4. Comparing different modes? Higher modes can have V_R both above and below the fundamental at a given frequency; always label the mode order.

    ──────────────────────────────────────────── 4. Quick estimator you can code in 2 lines (Python example)

    import numpy as np
    def v_rayleigh(vs, vp):           # Ewing et al. approximation
        beta = (vs/vp)**2
        return vs*(0.862+1.14*beta)
    
    # Example: Vs = 500 m/s, Vp = 1200 m/s (ν ≈ 0.31)
    print(v_rayleigh(500, 1200))      # → 461 m/s
    

    ──────────────────────────────────────────── Bottom line

    • In a single elastic half-space V_R lies ~4–13 % below V_S; the exact ratio is controlled by Poisson’s ratio (or V_P/V_S).
    • In layered ground, what you “expect” for V_R at a given frequency is whatever the depth-averaged V_S profile dictates. Always sanity-check against the homogeneous limits above, and label your plots as phase velocity unless you intentionally show group velocity.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.