The main reason of our ignorance of the nature of γ-ray burst (GRB) sources is the unfavourable combination of an unusual phenomenology and instrumental inadequacy. γ-ray telescopes have a poor imaging capability and GRBs only last from a fraction of a second to hundreds of seconds. In any case, after a short time they are no longer detectable in the γ-ray band even with large detectors. The burst decay is so fast and the positioning uncertainty so large that no search for delayed emission in other wavelengths has so far been successfully attempted7.

The Italian–Dutch Beppo-SAX satellite2,8 includes many experiments in different energy bands and with different fields of view. In particular, the combined presence of an all-sky Gamma-Ray Burst Monitor (GRBM)9,10 in the 40–700keV energy range, and two Wide Field Cameras (WFCs)11 in the 2–26keV energy range, which cover 5% of the sky with a pixel size of 5arcmin, allows an unprecedented capability of detecting and fast positioning GRBs and starting follow-up observations.

We developed a procedure for fast localization and rapid follow-up observations of GRBs with the Beppo-SAX Narrow Field Instruments (NFIs), a cluster of telescopes pointing towards the same field of view and covering the large band of 0.1–300keV (refs 10, 1214), taking advantage of having them aboard the same satellite and under the same Operation Control Centre.

On 1997 February 28.123620UTthe GRBM was triggered by a GRB event (GRB970228). When the data from the whole orbit were transferred to the ground station and forwarded to the Scientific Operation Centre, ‘quick-look’ analysis of data from the WFCs at the trigger time showed that a counting excess was also present in one WFC. The X-ray excess was imaged, showing a point-like source. WFC images before and after the event showed that the source was transient and simultaneous with the burst. Light curves in the γ-ray and X-ray band are shown in Fig. 1.

Figure 1: Time profile of GRB970228 in the γ-ray (from the Gamma-Ray Burst Monitor) and X-ray (from the Wide Field Camera) bands.
figure 1

The origin is the trigger time. The first pulse is shorter in γ-rays than in X-rays. Three other pulses follow (at 35, 50 and 70s from the trigger) that are much enhanced in the X-ray band. The total burst duration is 80s.

The burst position was first determined from a ‘quick-look’ analysis of the WFC data with an error radius of 10arcmin, suitable for planning a Target of Opportunity pointing (TOO1) of the GRB field with Beppo-SAX NFIs. After few hours, using off-line attitude analysis, we obtained for GRB970228 a refined error box of 3arcmin radius, centred at right ascension (RA) 05h01min57s, declination (dec.) 11°46.4′ (equinox 2000.0). With this refined position, observations in other wavelengths were solicited.

The first observation by the NFIs of Beppo-SAX started on February 28.4681, only 8 hours after the GRBM trigger, and ended on February 28.8330. The total exposure time was 14,344s in the Medium Energy Concentrator Spectrometer (MECS) and 8,725s in the Low Energy Concentrator Spectrometer (LECS). In the refined WFC error box we found only one source: 1SAX J0501.7+1146 with coordinates (equinox 2000.0) RA 05h01min44s, dec. 11°46.7′ and a 90% confidence error radius of 50arcsec.

As the pointing of NFIs was based on the first coarse positioning of the GRB in two of the three medium-energy telescopes, the source was partially covered by the window support structure. To exclude spurious variability due to pointing drifts, in the analysis we only use data from the LECS and only one out of three MECS units.

The source energy spectrum in the 0.1–10keV band is consistent with a power law of photon index 2.1 ± 0.3. The hydrogen column density is (3.5−2.3+3.3) × 1021cm−2 and consistent with the Galactic absorption along the line of sight 1.6 × 1021cm−2. The 2–10keV average source flux during this observation was (2.8 ± 0.4) × 10−12ergcm−2s−1, whereas the 0.1–2keV flux was (1.0 ± 0.3) × 10−12ergcm−2s−1 (note that the power-law photon index, the fluxes and the hydrogen column density quoted in ref. 5 are not correct). We also searched for hard X-ray emission (15–100keV) with the Phoswich Detection System without detecting any line or continuum flux. The 3σ upper limit on the 15–100keV emission is 4.3 × 10−11ergcm−2s−1, which is higher than the extrapolation from the low-energy power law.

We performed a second Target of Opportunity observation (TOO2) of the field with Beppo-SAX NFIs, about three days after the GRB970228 occurrence time (from March 3.7345 to March 4.1174). The exposure time was 16,270s with the MECS and 8,510s with the LECS. A source at a position consistent with that of 1SAXJ0501.7 + 1146 was detected in the MECS. Assuming the above spectral shape, the 2–10keV flux was (1.5 ± 0.5) × 10−13ergcm−2s−1, a factor 20 lower than in TOO1. The source was not detected in the LECS and the 3σ upper limit in the 0.1–2keV band was 4× 10−13ergcm−2s−1. In Fig. 2 we show the MECS image of the source in the first and in the second observation. This position is consistent with the GRB error box obtained with WFC, and with the GRB error annulus resulting from the Interplanetary Network (IPN) based on Beppo-SAX GRBM/Ulysses experiments15.

Figure 2: False-colour images of the source 1SAX J0501.
figure 2

7+1146, as detected in the error box of GRB970228 with Beppo-SAX Medium Energy Concentrator Spectrometer (2–10keV) during the first and second Target of Opportunity observations (TOO1 and TOO2, respectively). White corresponds to 31 counts per pixel2, green corresponds to 6 counts per pixel2and grey to a background of 0–1 counts per pixel2. Taking into account the correction for the number of telescopes (one in TOO1 and three in TOO2)and the vignetting in TOO1 due to off-axis pointing, the source faded by a factor of 20 in three days. From the ASCA faint sources data33, the probability that the source detected during the second pointing is coincident by chance with the position of 1SAX J0501,7 + 1146 is of the order of 1× 10−3.

No source was present in this position in the Rosat all-sky survey16 with a flux upper limit at 2.5σ of 1.9 × 10−13ergcm−2s−1, in the range 0.1–2.4keV, a value compatible with the LECS TOO2 but not with TOO1.

The transient time behaviour and the positional coincidence strongly support the association of 1SAX J0501.7 + 1146 with GRB970228. Using the statistics of X-ray sources derived from the GINGA background analysis17 we estimate that the probability to have by chance in a field of 3arcmin radius a source of intensity equal to or higher than the one we detected is <8 × 10−4. This probability value is reduced by a factor of at least 5 if we take into account the intersection of the error annulus of 30-arcsec half-width derived from IPN for GRB97022815,18 with the WFC and NFI error boxes.

Although results of a detailed spectral analysis of 1SAX J0501.7 + 1146 and GRB970228 will be reported elsewhere (F.Frontera et al., manuscript in preparation), we examine here the remarkable time behaviour of the source. Figure 3 shows the 2–10keV flux evolution during the two TOO observations. The source flux shows a significant decrease within the TOO1 observation. The reduced χ2(3 degrees of freedom, d.o.f.) for a constant flux is 3.6, corresponding to a probability of 0.13%. We tried to fit data of both observations with a single law. An exponential decay function does not fit the data. The best fit of the TOO1 and TOO2 flux data versus time was obtained with a power-law function (t−α) (see Fig. 3). The best-fit index is given by α= 1.33−0.11+0.132per d.o.f. = 0.7 with 4d.o.f.).

Figure 3: Variation of source flux with time in the 2–10keV range.
figure 3

Data from the TOO1 observation are grouped into four points of 8,000-s duration each. Data from TOO2 are grouped in one point only due to the lower statistics. The zero time is taken at the GRBM trigger time. Data are fitted by a power law (t−1.32). This law is shown as a solid sloping line at lower right. The forward extrapolation of the same law is consistent with the flux detected by ASCA34 on March 7.028 of (8 ± 0.3) × 10−14ergcm−2s−1 (averaged value for SIS and GIS detectors), in same energy range. The same law extrapolated backwards (dotted line) to the approximate time of the GRB (described by arrows in the top left) is a good match with the average flux of 2.3 × 10−8ergcm−2s−1 detected by WFC in the three minor pulses of Fig. 1 from 35 to 70s. Also shown is the 3σ upper limit of the source flux obtained with WFC5,000s after the burst, for an exposure time to the source of 1,000s.

We have also compared the flux and the decay law found for 1SAX J0501.7 + 1146 with the fluxes measured with GRBM and WFC during the γ-ray burst and during the following minor pulses shown in Fig. 1. In Fig. 3 (top left), the dashed line shows the 2–10keV flux averaged over 100s corresponding to the entire burst duration, whereas the solid horizontal line gives the average flux of the three minor pulses. Both fluxes are consistent with the backward extrapolation of the derived afterglow decay law. This strongly suggests that the X-ray emission detected soon after the GRB continuously evolves into the X-ray emission of the afterglow.

This result has an implication for the energetics of the event. The GRB fluence measured by GRBM in the 40–700keV band was 1.1 × 10−5ergcm−2. The X-ray fluence measured by WFC in the 2–10keV band was 1.2 × 10−6ergcm−2, that is 11% of the γ-ray fluence. If we assume that the three last pulses in Fig. 1 are part of the afterglow, by integrating the power law from 35s to infinity we find, in the window 2–10keV, a fluence which is 40% of the energy in the γ-ray burst itself in the band 40–700keV. The X-ray afterglow is not only the low-energy tail of the GRB phenomenon but is also a significant channel of energy dissipation of the event on a completely different timescale.

The well-established power-law decay function of the GRB remnant flux, the consistency of its extrapolation with the X-ray flux at the time of the burst, and the energetic content in X-rays are the main results of our discovery. They will significantly affect models of GRBs and constrain their parameters. Indeed the fast detection of GRB970228, promptly communicated to the scientific community3,4, triggered both the Beppo-SAX NFI follow-up and observations in the radio19,20 and optical bands21,22,23,24,25. These observations lead to cogent limits to the radio emission and to the detection5,26,27,28,29,30 of an optical transient, in a position consistent with that of 1SAXJ0501.7 + 1146 that faded in a few days. We note, however, that a previous GRB detected by Beppo-SAX, GRB97011131, had a γ-ray fluence about four times larger than GRB970228 and an undetectable X-ray emission 16hours after the burst. No fading optical source was detected at a level of magnitude B = 23 and R = 22.6 (ref. 32).

The Beppo-SAX measurement, in addition to discovering a relevant delayed X-ray emission, has thus provided the link missing for 25 years between the γ-ray phenomenology and the ultimate location capability of X-ray, optical and radio astronomy. We expect more detections of GRBs by Beppo-SAX GRBM/WFC, along with their follow-up observations. We hope that the existence of X-ray/optical afterglows and their rapid detection will contribute to the unambiguous identification of the GRB sources.