1 Introduction

The analysis of globular cluster systems in external galaxies is starting to fulfil its long-held promise as a probe of the formation of galaxies. Globular clusters (hereafter GCs) are a fossil record of this formation process, and provide one of the best tools with which to investigate the chemical enrichment and star formation history in the initial stages of galaxy formation (e.g., Ashman & Zepf 1998). Much recent interest has focused on the globular cluster systems of luminous elliptical galaxies where the combination of metallicities and kinematics can be used to distinguish between variants of the popular monolithic collapse and merger models for galaxy formation (e.g., Forbes et al. 1997; Sharples et al. 1998).

Lenticular galaxies hold a key position in the Hubble sequence of morphological types, intermediate between pure spheroidal systems like luminous ellipticals and disk-dominated spiral galaxies. Their formation mechanism is still the subject of considerable debate with evidence both for (Dressler et al. 1997) and against (Dressler 1980) their evolution from star-forming spirals via processes of gas stripping and exhaustion. A key question is when and how did such processes occur for S0 galaxies in a wide range of environments from rich clusters to the field. The globular cluster systems of S0 galaxies can provide independent constraints on when the major star formation episodes occurred both in the disk and halo. However, thus far they have been little studied with only NGC 1380 (Kissler-Patig et al. 1997) and NGC 4594 (the Sombrero galaxy, Bridges et al. 1997) having received any detailed attention and only with photometric methods.

NGC 3115 is one of the nearest S0 galaxies ($9.7\pm0.4$ Mpc, M_B=-20.1; Tonry et al. 2001) and is located in the sparse low-density environment of the Leo Spur. As such it provides an ideal test case for studying the formation mechanism of field S0's. A significant globular cluster system containing $\sim$500 clusters was first detected by Hanes & Harris (1986) using photographic plates. The nature of the cluster system and its origin were recently thrown into question with the discovery by Elson (1997) that the red-giant stars in the NGC 3115 halo $\sim$40 kpc from the centre showed a bimodal colour distribution. The inferred presence of two distinct halo populations of roughly equal size at metallicities of ${\rm [Fe/H]} \simeq -0.7$and ${\rm [Fe/H]} \simeq -1.3$ suggests at least two distinct epochs of formation.

The (V-I) colour distribution of the NGC 3115 globular cluster system has been the target of two recent independent studies using HST (Kundu & Whitmore 1998) and CFHT (Kavelaars 1998) data. Both studies find bimodality in the colour distributions of the GCs, with mean metallicities at ${\rm [Fe/H]} \simeq -0.37$ and ${\rm [Fe/H]} \simeq -1.36$ suggesting that the cluster and halo star systems may have formed coevally. This suggestion has gained further support from an investigation by Puzia et al. (2002b) who employed optical-IR colours to probe the globular cluster population close to the centre of NGC 3115. They also find two peaks in metallicity and an average age around $\approx$10 Gyr. However, their age discrimination power is very limited for metallicities lower than  ${\rm [Fe/H]} = -0.4$.

One scenario in which the above observations could be understood is if the metal-poor component corresponds to a primordial $\simeq$13 Gyr old population, whilst the metal-rich component formed a few Gyr later from enriched gas, possibly as the result of a minor merger (e.g., Bekki 1998). With only broad-band colours available, however, the well-known degeneracy between metallicity and age (Worthey 1994) makes such conclusions very uncertain.

For that reason, we have started a campaign to spectroscopically study the globular cluster system in NGC 3115. Our precision measurements of absorption line-strengths can be used to derive age and metallicity estimates directly from the comparison with new stellar population models. Unlike photometric methods, with spectroscopy we are also able to explore element abundance ratios for the GCs. We compare our results with other very recently obtained spectroscopic samples of GCs in early-type galaxies: in the giant Fornax elliptical NGC 1399 (Forbes et al. 2001), the Sa/Sb galaxy M 81 (Schroder et al. 2002), the SB0 galaxy NGC 1023 (Larsen & Brodie 2002) and the Sombrero galaxy NGC 4594 (Larsen et al. 2002).

The paper is organized as follows. In Sect. 2, the observations and their reduction are discussed. Section 3 presents the colour distribution of our sample while in Sect. 4 the treatment of abundance ratios and new stellar population models are investigated. Our results on abundance ratios, age and metallicity distributions for GCs in NGC 3115, the Milky Way (hereafter MW) and M 31 are presented in Sect. 5 with a general discussion in Sect. 6. We present our conclusions in Sect. 7.