Variable Persistent Emission Method

• Variable Persistent Emission Method is an X-ray analysis framework that allows the persistent emission normalization to vary during bursts using a free scaling factor (f_a).
• The method improves spectral fits by accounting for burst-driven changes in accretion flow, reducing residuals and providing insights into disk and corona behavior.
• It has been validated in thermonuclear bursts, superbursts, and magnetar studies, demonstrating its utility in both fixed-shape spectral models and RMS-based variability analyses.

Searching arXiv for the core papers on the variable persistent emission method and closely related burst spectroscopy applications. Variable persistent emission method denotes an X-ray timing and spectral-analysis framework in which the persistent component is not assumed to remain fixed during intervals traditionally treated as burst-only emission. In thermonuclear-burst spectroscopy, the method replaces subtraction of an immutable pre-burst spectrum with a model in which the persistent contribution is multiplied by a free scaling factor $f_a$ or, in Comptonization-based implementations, by a time-dependent normalization. In magnetar analyses, a related formulation quantifies over-Poisson variability in the persistent light curve through an RMS statistic and interprets that variability with a micro-burst model. Across these uses, the common methodological move is to elevate persistent emission from a fixed background term to an explicitly variable observable (Worpel et al., 2015, Nakagawa et al., 2018).

1. Canonical $f_a$ formulation in thermonuclear-burst spectroscopy

In the burst-spectroscopy formulation, let $P(E)$ be the best-fit model to the pre-burst persistent spectrum, including Galactic absorption; let $B(E;T_{\rm bb},K_{\rm bb})$ be a simple blackbody describing the burst emission; and let $b_{\rm inst}(E)$ denote instrumental background. The variable-persistent-emission model is

$$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$

with $A(E)=\exp[-n_H\,\sigma(E)]$. In shorthand, the method is often written as

$$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$

Within this formalism, $f_a$ quantifies whether accretion-powered emission brightens $(f_a>1)$ or dims $f_a$0 during the nuclear flash. The standard approach for time-resolved X-ray spectral analysis of thermonuclear bursts instead subtracts the entire pre-burst emission as background and fits the residual burst spectrum with a single blackbody, thereby assuming both a constant persistent spectral shape and a persistent normalization fixed exactly at the pre-burst level. Introducing $f_a$1 relaxes only the normalization constraint: the shape of $f_a$2 remains frozen, but its intensity may vary on burst timescales (Worpel et al., 2015).

This distinction is methodologically important. Under the standard subtraction approach, any real burst-driven change in disk or coronal emission is forced into the burst blackbody fit or into residual structure. Under the $f_a$3 approach, such variability is absorbed into an explicit parameter. A plausible implication is that the method functions simultaneously as a fitting improvement and as a diagnostic of burst–accretion-flow coupling.

2. Fitting workflow, parameter control, and statistical validation

The procedure begins with pre-burst modeling. For each burst, a 16 s pre-burst spectrum is accumulated and fitted with a suite of candidate XSPEC models, such as absorbed disk plus Comptonization forms, with the model of minimum $f_a$4 chosen to define $f_a$5. Time-resolved burst spectroscopy then proceeds by dividing the burst into short intervals, for example $f_a$6 s to a few seconds, while ensuring $f_a$7 counts. Each interval is fitted with the composite model $f_a$8, keeping the shape of $f_a$9 fixed and allowing $P(E)$0, $P(E)$1, and $P(E)$2 to vary freely. A uniform prior is effectively adopted for $P(E)$3 over a broad range such as $P(E)$4 to $P(E)$5, although physically one typically enforces $P(E)$6 (Worpel et al., 2015).

Model selection is performed by comparing goodness-of-fit with and without a free $P(E)$7. A typical $P(E)$8 for one additional degree of freedom indicates significant improvement, and in the RXTE sample the Bayes factor favoring the variable-$P(E)$9 approach over the standard method is approximately $B(E;T_{\rm bb},K_{\rm bb})$0, even after penalizing the extra parameter. Procedural validation includes checking that $B(E;T_{\rm bb},K_{\rm bb})$1 returns to approximately unity in pre-burst and late-tail intervals. Practical recommendations include extracting a high-quality pre-burst spectrum with at least $B(E;T_{\rm bb},K_{\rm bb})$2 counts, fitting multiple plausible persistent models, using sufficiently short time bins to follow rapid $B(E;T_{\rm bb},K_{\rm bb})$3 evolution while retaining $B(E;T_{\rm bb},K_{\rm bb})$4, and adopting thresholds such as $B(E;T_{\rm bb},K_{\rm bb})$5 for one extra parameter at $B(E;T_{\rm bb},K_{\rm bb})$6 confidence when claiming a significant $B(E;T_{\rm bb},K_{\rm bb})$7 deviation (Worpel et al., 2015).

The central statistical point is that the method does not merely add flexibility. It tests a specific null hypothesis—constant persistent normalization—and evaluates whether the data justify replacing that null by a variable normalization while preserving the pre-burst spectral shape.

3. Empirical behavior of $B(E;T_{\rm bb},K_{\rm bb})$8 and its physical interpretation

A large RXTE reanalysis of $B(E;T_{\rm bb},K_{\rm bb})$9 photospheric radius expansion bursts from $b_{\rm inst}(E)$0 sources found that, for the majority of spectra, the best-fit value of $b_{\rm inst}(E)$1 is significantly greater than $b_{\rm inst}(E)$2, indicating that the persistent emission typically increases during a burst. Elevated $b_{\rm inst}(E)$3 values were measured not only during the radius-expansion interval but also in the cooling tail. The modified model yields a lower average value of the $b_{\rm inst}(E)$4 fit statistic, although not yet to the level of formal statistical consistency for all spectra. In the same study, an inverse correlation of $b_{\rm inst}(E)$5 with the persistent flux was measured, consistent with theoretical models of disk response (Worpel et al., 2015).

In the broader burst sample summarized for the method, typical peak values of $b_{\rm inst}(E)$6 in PRE bursts reach approximately $b_{\rm inst}(E)$7–$b_{\rm inst}(E)$8 times the pre-burst level, while in non-PRE bursts they commonly rise by factors of $b_{\rm inst}(E)$9–$$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$0. The characteristic time profile begins near $$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$1 before the burst, rises during the burst rise, sometimes peaks around photospheric touchdown, and then decays back toward unity in the cooling tail. Non-PRE bursts show a strong positive correlation between instantaneous burst flux $$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$2 and $$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$3, with Kendall $$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$4 at $$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$5 in $$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$6 of cases. When normalized by the persistent-to-Eddington ratio $$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$7, the product $$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$8 is empirically bounded above by $$S(E)=A(E)\cdot B(E;T_{\rm bb},K_{\rm bb})+f_a\cdot P(E)+b_{\rm inst}(E),$$9 for PRE bursts and $A(E)=\exp[-n_H\,\sigma(E)]$0 for non-PRE bursts (Worpel et al., 2015).

The usual physical interpretation is Poynting–Robertson drag. During a bright burst, intense radial photon flux from the neutron-star surface can remove angular momentum from inner-disk material and transiently raise the mass flow onto the star. In this reading, the instantaneous accretion rate is written as

$A(E)=\exp[-n_H\,\sigma(E)]$1

and simple analytic estimates relate $A(E)=\exp[-n_H\,\sigma(E)]$2. At very high burst luminosity approaching $A(E)=\exp[-n_H\,\sigma(E)]$3, the disk may be temporarily evacuated or partially disrupted, so the observed spectral signatures can be more complex than a pure normalization change. The method therefore supports, but does not uniquely prove, an accretion-rate interpretation (Worpel et al., 2015).

4. Extensions to superbursts and instrument-specific implementations

During the 2021 superburst of 4U 1820–30, time-resolved spectra from NICER and MAXI were modeled with $A(E)=\exp[-n_H\,\sigma(E)]$4, and in some tail intervals with $A(E)=\exp[-n_H\,\sigma(E)]$5. Here $A(E)=\exp[-n_H\,\sigma(E)]$6 in disk geometry carries free parameters $A(E)=\exp[-n_H\,\sigma(E)]$7, $A(E)=\exp[-n_H\,\sigma(E)]$8, $A(E)=\exp[-n_H\,\sigma(E)]$9, and a normalization $$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$0; the normalization serves as the tracer of variable persistent emission and is identified with the bolometric Comptonization flux $$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$1. NICER burst-tail spectra were extracted in $$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$2 s bins over $$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$3–$$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$4 keV, with $$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$5 errors derived by the $$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$6 criterion. The recovered persistent flux followed a logistic form,

$$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$7

with $$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$8, $$\text{total\_spectrum}=f_a\times \text{persistent\_spectrum}+\text{burst\_spectrum}.$$9, and $f_a$0. The associated $f_a$1–$f_a$2 rise time is approximately $f_a$3 hr. The minimum persistent flux at the superburst peak was estimated as $f_a$4, implying $f_a$5, described as nearly complete quenching. Comparison of the superburst total energy, $f_a$6 erg, with the gravitational binding energy of disk material between $f_a$7 and $f_a$8, $f_a$9 erg, suggests that radiation pressure or Poynting–Robertson drag can evacuate the inner disk; the subsequent recovery timescale is consistent with standard $(f_a>1)$0-disk viscous times of $(f_a>1)$1–$(f_a>1)$2 hr for $(f_a>1)$3–0.2 and $(f_a>1)$4 (Peng et al., 2024).

The same event also exhibited a transient absorption line that shifted from $(f_a>1)$5 to $(f_a>1)$6 keV in the $(f_a>1)$7–$(f_a>1)$8 hr interval. Assigning it to Ar XVIII K$(f_a>1)$9 with rest energy $f_a$00 keV gives a gravitational redshift

$f_a$01

and hence

$f_a$02

For $f_a$03 keV, one obtains $f_a$04 and $f_a$05 km for $f_a$06. The absorption feature was interpreted as likely originating in the inner accretion disk rather than in burst emission from the neutron-star surface, and its evolution suggested inward recovery of the disk (Peng et al., 2024).

A NuSTAR study of 4U 1323–62 provides a source-specific $f_a$07 implementation using a pre-burst persistent model $f_a$08 and a burst model $f_a$09. In the pre-burst fit, $f_a$10 was frozen, the absorption edge was found at $f_a$11 keV with $f_a$12, and the Comptonization parameters were $f_a$13, $f_a$14 keV, and seed-$f_a$15 keV. During burst fits, all persistent-shape parameters were frozen, while only $f_a$16, burst $f_a$17, and burst normalization were allowed to vary over $f_a$18–$f_a$19 keV. For three bursts divided into five segments $f_a$20–$f_a$21 of $f_a$22 s, except a final $f_a$23 s segment, all three showed $f_a$24 rising from a pre-burst value near unity to a maximum in $f_a$25, then declining toward quiescence. The largest reported enhancement was $f_a$26 in burst B2. In that study, the method improved the fit by $f_a$27 with only one extra parameter over a simple blackbody model, recovered average apparent blackbody radii of $f_a$28–$f_a$29 km, and yielded $f_a$30 values spanning approximately $f_a$31–$f_a$32 (Bhattacharya et al., 5 Nov 2025).

5. RMS-based persistent-emission variability in magnetars

In magnetar work, the expression “variable persistent emission” denotes a different but related methodology. For a background-subtracted light curve with counts $f_a$33, counting errors $f_a$34, and $f_a$35 bins, the dimensionless RMS intensity variation is defined by

$f_a$36

where $f_a$37. The subtraction of $f_a$38 removes the variance expected from counting statistics, so the residual RMS measures intrinsic over-Poisson source variability. Nakagawa et al. proposed that the persistent X-ray emission of magnetars is the superposition of numerous short, $f_a$39 ms micro-bursts of various fluences $f_a$40, with cumulative number–fluence relation

$f_a$41

where $f_a$42 from observations of strong bursts, over $f_a$43 with $f_a$44 and $f_a$45. The associated probability density is

$f_a$46

and the expected fractional RMS due solely to micro-burst statistics is

$f_a$47

Inserting $f_a$48, the quoted $f_a$49 and $f_a$50, and a typical persistent X-ray flux of $f_a$51, implying $f_a$52 in $f_a$53–$f_a$54 keV, gives $f_a$55, consistent with observed values of approximately $f_a$56–$f_a$57 (Nakagawa et al., 2018).

The observational implementation used Suzaku XIS $f_a$58–$f_a$59 keV light curves with $f_a$60 s bins, or $f_a$61 s for multi-band analysis, and HXD-PIN $f_a$62–$f_a$63 keV light curves with $f_a$64 s bins. Bright bursts were removed by flagging bins above $f_a$65, visually verifying them, and recomputing the RMS as $f_a$66. Across $f_a$67 magnetars and $f_a$68 observations, significant excess RMS intensity variations were found in all $f_a$69 objects. In four magnetars, corresponding to six observations, the RMS increased clearly toward higher energy bands; in those cases $f_a$70 rose above the soft–hard crossover of approximately $f_a$71–$f_a$72 keV and tracked the hard power-law component rather than the thermal blackbody. The authors interpreted these results as evidence that persistent emission and burst emission have identical emission mechanisms and that the soft thermal component and hard X-ray component are emitted from different regions far apart from each other. Monte Carlo checks further indicated that spin modulation on $f_a$73–$f_a$74 s periods and day-scale flux drifts do not bias the RMS when binning of at least $f_a$75 s is used (Nakagawa et al., 2018).

6. Advantages, limitations, and interpretive boundaries

The principal advantage of the method in burst spectroscopy is improved fit fidelity. Allowing $f_a$76 to vary lowers the average $f_a$77 statistic relative to the standard background-subtraction approach, and in the NuSTAR application it removed high-energy residuals while improving the fit by $f_a$78 with only one additional parameter. It also yields more reliable blackbody temperatures and radii because the persistent continuum is no longer forced into the burst tail residuals (Worpel et al., 2015, Bhattacharya et al., 5 Nov 2025).

Its chief limitation is structural: the method assumes that the shape of the persistent spectrum does not change, only its normalization. Rapid coronal cooling, observed above approximately $f_a$79 keV, or disk ionization changes could violate that assumption. Within the approximate $f_a$80–$f_a$81 keV PCA band and on timescales $f_a$82 s, however, no significant shape changes were detected outside bursts. Parameter degeneracy is another persistent concern. $f_a$83 is often strongly covariant with the blackbody normalization, leading to larger uncertainties in both; if the persistent spectrum closely mimics a blackbody, $f_a$84 may be poorly constrained. The method is also sensitive to the quality of the pre-burst fit, so errors in $f_a$85, seed temperature, edge energy, or continuum choice can propagate directly into the inferred burst parameters (Worpel et al., 2015, Bhattacharya et al., 5 Nov 2025).

Interpretively, $f_a$86 should not automatically be identified with a pure change in accretion rate. The physical reading $f_a$87 is explicitly conditional on other processes—such as reflection, corona collapse, or more general changes in the Comptonizing medium—being ruled out or modeled. The superburst application strengthens the case that persistent emission can also decrease dramatically, since the Comptonization component in 4U 1820–30 was inferred to be nearly completely quenched before recovering on a viscous timescale (Peng et al., 2024). In magnetar work, similarly, excess persistent-emission variability is not attributed to counting noise because the Poisson term is explicitly subtracted and the residual RMS is tested against instrumental and timing-systematics checks (Nakagawa et al., 2018).

Taken together, these studies establish variable persistent emission as a methodological category rather than a single code path. In bursting low-mass X-ray binaries it is primarily a spectral-decomposition strategy centered on $f_a$88 or its Comptonization-normalization analogue; in magnetars it is an RMS-based variability formalism tied to a micro-burst hypothesis. What unifies these uses is the rejection of a strictly static view of persistent X-ray emission during high-energy activity.