Thamnos: Stellar-Halo Substructure

• Thamnos is a stellar-halo substructure characterized by low-energy retrograde motion and distinct dynamical properties, likely representing ancient accreted debris.
• It has been identified using a variety of clustering methods in orbital, chemical, and kinematic phase spaces, highlighting both genuine accreted stars and contaminant populations.
• Chemical studies reveal a bimodal abundance pattern that differentiates a metal-poor accreted component from contributions associated with GSE and ω Cen, underscoring Thamnos’ complex formation history.

Thamnos is a stellar-halo substructure of the Milky Way, identified in the Gaia era as a low-energy retrograde component of the local halo and generally interpreted as debris from an ancient accretion event (Koppelman et al., 2019). Across the subsequent literature, Thamnos has been recovered in multiple dynamical and chemo-dynamical searches, but its astrophysical status remains contested: some studies treat it as a distinct low-mass progenitor, while others argue that the usual “Thamnos” selection region in integral-of-motion space is a superposition of genuine accreted debris, Gaia-Sausage-Enceladus (GSE), heated in situ halo or disc stars, and possibly $\omega$ Cen-related material (Mori et al., 16 Sep 2025).

1. Discovery and basic dynamical identity

Thamnos was first isolated in a local halo sample built from Gaia DR2, supplemented with RAVE, APOGEE, and LAMOST, by applying HDBSCAN in the four-dimensional space of specific orbital energy $E$, angular momentum $L_z$, orbital eccentricity, and metallicity (Koppelman et al., 2019). In that work, two overlapping clumps were linked and collectively named Thamnos. The defining dynamical quantities were

$$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$

In the original characterization, Thamnos occupied a more tightly bound region than Sequoia, with Thamnos 1 at $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$ and circularity $< -0.75$, and Thamnos 2 at $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$ and $-0.75<\mathrm{circ}< -0.40$ (Koppelman et al., 2019). The substructure was described as retrograde, low-inclination, and mildly eccentric, with $$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$ for Thamnos 2 and $\simeq -200\,\mathrm{km\,s^{-1}}$ for Thamnos 1, and with eccentricities $E$0–$E$1 (Koppelman et al., 2019).

Later analyses recovered the same component in larger local-halo samples. In a very metal-poor Gaia DR3 + LAMOST DR9 sample, Thamnos was identified in action-energy space with mean orbital parameters $E$2, $E$3, $E$4, and $E$5 (Ye et al., 2023). A separate LAMOST + Gaia study found a larger Thamnos sample with mean eccentricity $E$6 and mean $E$7, again reinforcing the picture of a moderately eccentric, low-energy retrograde population (Deepak, 2024).

2. Selection functions and identification methodologies

Although the physical interpretation of Thamnos differs across papers, its operational definition is usually tied to a bounded region of orbital phase space. A common APOGEE-based selection adopts

$E$8

which yielded $E$9 stars in one chemical-characterization study and was reused in later contamination analyses (Horta et al., 2022, Mori et al., 16 Sep 2025).

Different groups have recovered Thamnos with distinct clustering formalisms. These include Shared Nearest Neighbor clustering in $L_z$0 space (Ye et al., 2023), the StarGO self-organizing-map algorithm in the seven-dimensional space $L_z$1 (Li et al., 2 Apr 2025), t-SNE chemo-kinematic tagging based on $L_z$2 (Youakim et al., 30 Oct 2025), wavelet transforms in $L_z$3 space (Kushniruk et al., 23 Feb 2026), CLiMB novelty detection for RR Lyrae in $L_z$4 (Muraveva et al., 23 Feb 2026), and GS3 Hunter, which combines Siamese neural networks with K-means (Wang et al., 21 Dec 2025).

Representative definitions illustrate both the convergence and the non-uniformity of the literature:

Study	Phase-space definition	Reported outcome
Koppelman et al.	Thamnos 1: $L_z$5, $L_z$6; Thamnos 2: $L_z$7, $L_z$8	Two low-energy retrograde clumps (Koppelman et al., 2019)
APOGEE chemical characterization	$L_z$9, $$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$0, $$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$1	$$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$2 stars (Horta et al., 2022)
GALAH DR4 wavelet analysis	Thamnos 1 centered at $$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$3, $$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$4; Thamnos 2 at $$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$5, $$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$6	Two retrograde groups (Kushniruk et al., 23 Feb 2026)
RR Lyrae CLiMB analysis	Semi-supervised ellipsoidal assignment in $$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$7 around labeled Thamnos centroids	Sample increased from $$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$8 to $$E = \tfrac12\,(v_r^2+v_\theta^2+v_\phi^2)+\Phi(r), \qquad L_z = R\,v_\phi.$$9 RR Lyrae (Muraveva et al., 23 Feb 2026)

The sign convention for retrograde motion is not uniform across all summaries. Some papers use the standard $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$0 or $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$1 for retrograde motion, whereas others report the same substructure with opposite-sign conventions for $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$2 or $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$3 (Koppelman et al., 2019, Woody et al., 2024). This suggests that cross-study comparisons of Thamnos must track the adopted convention explicitly rather than rely on sign alone.

3. Chemical-abundance patterns and metallicity structure

The chemical characterization of Thamnos is one of the main reasons it has remained astrophysically significant. In APOGEE-based analyses, Thamnos typically appears as a metal-poor, $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$4-enhanced population distinct from GES and from disc-like structures. One study reported $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$5 with a range from $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$6 to $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$7, together with $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$8, $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.65,-1.45]$9, $< -0.75$0, $< -0.75$1, and $< -0.75$2 (Horta et al., 2022). In the same work, comparison with GES at $< -0.75$3 gave $< -0.75$4 with $< -0.75$5, supporting chemical distinctness (Horta et al., 2022).

A separate high-resolution study of low-$< -0.75$6 retrograde groups overlapping Thamnos found a more metal-poor dominant population. Its metallicity distribution was fit by two Gaussians with peaks at $< -0.75$7 and $< -0.75$8, where the lower-metallicity component was identified as the main Thamnos population (Xie et al., 14 Jan 2026). The same analysis reported a flat $< -0.75$9-plateau with $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$0, $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$1, $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$2, and $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$3, together with no evidence for an $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$4 knee (Xie et al., 14 Jan 2026).

Recent GALAH DR4 work further resolved the retrograde low-energy region into Thamnos 1 and Thamnos 2. In that analysis, kernel-density estimates gave metallicity peaks of $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$5 for Thamnos 1 and $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$6 for Thamnos 2, while median abundance differences indicated higher $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$7 for Thamnos 2, with approximately $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$8 dex versus $E/10^5\,(\mathrm{km^2\,s^{-2}})\in[-1.80,-1.60]$9 dex for Thamnos 1 (Kushniruk et al., 23 Feb 2026). Thamnos 2 also showed stronger $-0.75<\mathrm{circ}< -0.40$0-process signatures, with $-0.75<\mathrm{circ}< -0.40$1 versus $-0.75<\mathrm{circ}< -0.40$2 for Thamnos 1 (Kushniruk et al., 23 Feb 2026).

Taken together, these results suggest two overlapping chemical descriptions of Thamnos. One is a kinematically selected envelope centered near $-0.75<\mathrm{circ}< -0.40$3 to $-0.75<\mathrm{circ}< -0.40$4 with moderate $-0.75<\mathrm{circ}< -0.40$5 enhancement; the other is a very metal-poor component near $-0.75<\mathrm{circ}< -0.40$6 that several later studies interpret as the genuine accreted core. This interpretation is strengthened by the repeated claim that no clear $-0.75<\mathrm{circ}< -0.40$7 knee is seen in the chemically cleaner, metal-poor component (Horta et al., 2022, Xie et al., 14 Jan 2026).

4. Stellar ages, star-formation history, and progenitor chronology

Chronological studies consistently place Thamnos among the oldest identified halo building blocks, though the exact age scale depends on tracer selection and fitting methodology. In CMD-fitting work based on a local 5D Gaia DR3 sample, the stricter “Thamnos B” selection yielded a metallicity distribution extending from approximately $-0.75<\mathrm{circ}< -0.40$8 to $-0.75<\mathrm{circ}< -0.40$9 dex and a half-mass lookback time

$$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$0

with a steep early star-formation history that rapidly declined for younger ages (Dodd et al., 2024).

A complementary H3 survey analysis of MSTO and subgiant stars reached an even earlier chronology. For 35 stars classified as Thamnos members, nearly all inferred ages exceeded $$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$1, with a sharp peak near $$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$2 (Woody et al., 2024). Fitting a truncated-Gaussian star-formation history gave $$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$3, $$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$4, and $$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$5, with $$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$6 of the stellar mass assembled by $$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$7 (Woody et al., 2024). Under the assumption that SFH truncation marks accretion, that study inferred $$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$8 and described Thamnos as the earliest identified accretion event in the metal-poor halo (Woody et al., 2024).

A broader age compilation based on LAMOST and Gaia found somewhat younger but still ancient values: mean age $$\langle v_\phi\rangle \simeq -150\,\mathrm{km\,s^{-1}}$$9, mode $\simeq -200\,\mathrm{km\,s^{-1}}$0, and peak metallicity near $\simeq -200\,\mathrm{km\,s^{-1}}$1 dex for a 3006-star Thamnos sample (Deepak, 2024). This does not contradict the older determinations so much as indicate method dependence in age recovery and in sample contamination.

The progenitor mass inferred for Thamnos is generally small. The discovery paper estimated $\simeq -200\,\mathrm{km\,s^{-1}}$2 from mock dwarf contours in $\simeq -200\,\mathrm{km\,s^{-1}}$3–$\simeq -200\,\mathrm{km\,s^{-1}}$4 space (Koppelman et al., 2019), and later high-resolution work continued to describe the genuine accreted component as a tiny, metal-poor dwarf with $\simeq -200\,\mathrm{km\,s^{-1}}$5 (Ceccarelli et al., 7 Oct 2025). A plausible implication is that Thamnos represents an early, low-mass accretion event whose star formation was truncated quickly after infall.

5. Internal complexity, contamination, and relation to other systems

The main controversy surrounding Thamnos is whether it is a chemically coherent relic of a single minor merger or a mixed region in integral-of-motion space. Several recent studies favor the latter interpretation. A chemically driven Gaussian-mixture-model comparison of kinematically selected halo substructures found that Thamnos likely contains GSE and heated disc stars in significant amounts (Mori et al., 16 Sep 2025). In that analysis, the fraction chemically compatible with GSE was reported as $\simeq -200\,\mathrm{km\,s^{-1}}$6 for one definition, corresponding to an absolute $\simeq -200\,\mathrm{km\,s^{-1}}$7, or $\simeq -200\,\mathrm{km\,s^{-1}}$8 under an alternative definition; the metal-poor-disc-compatible fraction was $\simeq -200\,\mathrm{km\,s^{-1}}$9; and among the remaining outliers, $E$00 aligned chemically with $E$01 Cen (Mori et al., 16 Sep 2025).

Dedicated UVES spectroscopy sharpened the same point. In a sample of 140 Thamnos candidates, the metallicity distribution was best fit by two Gaussians: a “true Thamnos” component with mean $E$02 and $E$03, comprising $E$04 of the area, and a contaminant component with mean $E$05 and $E$06, comprising $E$07 of the area (Ceccarelli et al., 7 Oct 2025). That work concluded that the dynamically selected Thamnos region is dominated by in situ halo stars, with a smaller metal-poor accreted component and residual GSE contamination below $E$08 (Ceccarelli et al., 7 Oct 2025).

RR Lyrae analyses point in the same direction. In a CLiMB-based Gaia DR3 study, the Thamnos RR Lyrae metallicity distribution was explicitly bimodal: a metal-poor peak with $E$09 stars at $E$10 dex and $E$11 dex, and a metal-rich peak with $E$12 stars at $E$13 dex and $E$14 dex (Muraveva et al., 23 Feb 2026). The metal-poor peak was interpreted as the genuine Thamnos population, whereas the metal-rich peak was attributed to contamination from GSE and Splash/Aurora-like stars (Muraveva et al., 23 Feb 2026).

Relations to $E$15 Cen have also become part of the modern discussion. A chemo-kinematic tagging study recovered a small Thamnos cluster containing NGC 288, NGC 5139 ($E$16 Cen), the NGC 288 stream, and the Fimbulthul stream, and argued that the chemo-dynamic properties of $E$17 Cen are consistent with a common accretion with Thamnos (Youakim et al., 30 Oct 2025). This does not establish identity between the two systems, but it does reinforce the idea that the standard Thamnos region may host multiple lineages of accreted debris.

An earlier chemodynamical study had already hinted at a link between Thamnos and GSE. In a broad retrograde sample, stars selected to probe Thamnos and GSE were different in overall metallicity and $E$18, but shared the same radial and vertical metallicity gradients, the same positive $E$19, and the same eccentricity-metallicity trend (Kordopatis et al., 2020). This suggests that even if Thamnos is a distinct structure, its observed local manifestation overlaps dynamically with debris from other early halo-building events.

6. Specialized tracers and uncommon abundance signatures

Beyond ordinary giant-star samples, Thamnos has been traced with chemically distinctive populations. In the R-Process Alliance extended sample of $E$20-process-enhanced halo stars, Thamnos was re-identified as CDTG-22, containing $E$21 member stars with $E$22, $E$23, and $E$24 (Shank et al., 2022). That work interpreted Thamnos as a chemically old, $E$25-process self-enriched system, consistent with early enrichment before disruption (Shank et al., 2022).

An even more unusual result concerns Thamnos-2 and beryllium. A recent abundance study of nine stars associated with Thamnos found four new Be-rich members and showed that the previously known Be-rich stars HD 106038 and HD 132475 are also consistent with Thamnos membership (Molaro et al., 9 Jul 2026). In that sample, all currently known Be-rich stars were associated with Thamnos-2, and the Be enhancement correlated tightly with silicon and neutron-capture elements, including

$E$26

and

$E$27

(Molaro et al., 9 Jul 2026). The authors interpreted this pattern as the imprint of a rare spallation event, plausibly associated with a hypernova in the Thamnos-2 progenitor (Molaro et al., 9 Jul 2026).

Thamnos has also been extended into the substellar regime. A survey of cold subdwarfs reported that the extreme L subdwarf 2MASS J053253.46+824646.5 and the mild T subdwarf CWISE J113010.07+313944.7 may be part of the Thamnos population on the basis of full 3D LSR velocities: for J0532+8246, $E$28, $E$29, $E$30; for J1130+3139, $E$31, $E$32, $E$33 (Burgasser et al., 2024). This broadens the empirical scope of Thamnos beyond traditional stellar tracers.

The cumulative literature therefore presents Thamnos as both a named substructure and an evolving interpretive problem. Its repeated recovery in independent surveys supports the reality of a low-energy retrograde halo component, while its chemically resolved bimodality and cross-contamination with GSE, in situ halo populations, and possibly $E$34 Cen-related debris indicate that “Thamnos” often denotes a structured region of phase space rather than a single mono-chemical population (Koppelman et al., 2019, Ceccarelli et al., 7 Oct 2025). The strongest common ground across recent work is that a genuinely ancient, metal-poor accreted component is present within that region, and that this component records one of the earliest low-mass mergers contributing to the Milky Way’s inner halo (Dodd et al., 2024).