betaFeedback

WorksPhysics and Astronomy

Near-exact Upper Confidence Limits for Axion Haloscopes

Abstract

The data acquisition of axion haloscopes commonly implements sequential adaptive frequency scans. We present a method to calculate likelihood-based upper confidence limits on axion-photon coupling under this circumstance. We compare the likelihood-based method conceptually and quantitatively with tailor-made methods, i.e. the one proposed by Brubaker et al. (Geometric) and its upgrade the Bayesian Power-Measured (BPM) framework proposed by Palken et al. We find that the likelihood method yields average upper limits that are significantly more constraining than those obtained from these methods. For example, expected constraints improve by a factor of 1.4 as compared with the current state of the art when applied to projections for the ALPHA Plasma-haloscope. We further clarify the statistical basis for the Geometric and BPM methods. In particular, we point out the correspondence of BPM to e-values, proposed as an alternative to conventional p-values in statistics.

Introduction

The nature of dark matter is one of the most pressing problems in modern fundamental physics [31, 30, 29, 28, 27, 26, 25]. Among the many particle candidates, axions are of particular interest. Their existence is motivated by symmetry considerations in the strong interactions [39, 38, 37, 36, 35, 34, 33, 32], and well motivated production mechanisms imply that axions could constitute dark matter[22, 23, 24]. Possible axion masses span 19 orders of magnitude, so an abundance of experimental techniques is applied to find them [15, 16, 17, 18, 20, 21, 19].

A particularly promising venue is provided by “haloscopes” [40, 41]—cavities that can detect the power deposited by an axion conversion to photons prompted by a strong magnetic field. Haloscopes operate by scanning a frequency range for a narrow-band excess at an unknown frequency νa\nu_a, corresponding to the axion mass ma≈hνa/c2m_a\approx h \nu_a/c^2. Haloscope analysis protocols [e.g., 5, 6, 7, 14, 8, 13, 9] have the possibility of acquiring additional data at promising frequencies. A power excess that meets a certain threshold condition triggers one (or more) additional measurements, also called “rescans.” This is a very useful feature for axion searches, but poses also a challenge to statistical inference, as data acquisition is adaptive and conditioned on the observed data itself.

Upper Limits in axion haloscopes

In haloscopes, the observed excess XνX_{\nu} at a frequency ν\nu is Gaussian with very good approximation [6, 8]. Its mean is null if no axion is present; otherwise, it is proportional to the axion-photon coupling, henceforth generally denoted as A\mathbb{A} (or AA) for the true (or tested) value:

Xν∼N(Auν,uν).\begin{equation} X_\nu\sim\mathcal{N}(\mathbb{A}u_\nu, \sqrt{u_\nu}). \tag*{(1)} \end{equation}

Here, uν>0u_{\nu} > 0 is an alias for the signal-to-noise ratio at ν\nu, given the specific experimental setting and measurement time (see also [9]). With this notation, the (i+1i+1)th rescan is triggered whenever Xν,iX_{\nu, i} satisfies Xν,i≥bν,iX_{\nu,i}\ge b_{\nu,i}, where bν,ib_{\nu,i} is a threshold commonly defined in terms of the system noise. After assessing the presence of an axion signal (rejection of noise-only hypothesis) anywhere in the frequency range, statistical inference turns to exclusion. It consists of the point-wise statement on the maximal coupling A\mathbb{A} at a given frequency ν\nu still consistent with the observation at a certain confidence level (CL). Since exclusion is defined pointwise in frequency, we will omit the subscript ν\nu from now on.

In the former case the null hypothesis is noise, in the latter it is \mathbb{A} > A_{\text{\textsc{ul}}}. Conflating the two can lead to unwanted results. For example, the protocol implemented by Brubaker et al. and used in early HAYSTAC analysis [5, 6] applies a geometric test with binary outcome (signal/no-signal) defined by the local 5σ5\sigma significance for the false-negative rate of a given discovery benchmark A_{\text{\textsc{t}}}. The latter corresponds to the prescribed CL at each stage thanks to a calibration of the uiu_i values. The same A_{\text{\textsc{t}}} becomes the quoted value for exclusion when no evidence is found, and the same nominal exclusion CL is assigned to candidates failing any of the scans.

However, this argument conditions on passing the previous scan(s). A frequentist exclusion must be calibrated over the full decision tree (i.e., all possible experimental outcomes) whose total false negative probability is Fn=f1+(1−f1)f2F_n = f_1+(1-f_1)f_2, with fi=Φ[(bi−Aui)/ui]f_i = \Phi[(b_i - Au_i)/\sqrt{u_i}], where Φ\Phi is the standard normal CDF.

In general, the geometric upper limit A_{\text{\textsc{ul}}}^{g_{N_s}} satisfies:

\begin{equation} \alpha = \sum_{i=1}^{N_s}\frac{f_i(A_{\text{\textsc{ul}}}^{g_{N_s}})}{1-f_i(A_{\text{\textsc{ul}}}^{g_{N_s}})} \prod_{k=1}^{i}\left[1-f_k(A_{\text{\textsc{ul}}}^{g_{N_s}})\right], \tag*{(2)} \end{equation}

where NsN_s is the maximum number of allowed scans. Notably, definition (2) implies that the upper limit weakens as NsN_s increases, since it raises the cumulative false-negative probability. Moreover, while its binary nature protects against downward fluctuations, it ignores information from previous scans. Hence, the result becomes overly conservative and closer to an a priori sensitivity benchmark.

In addition, exclusion (2) and frequentist confidence intervals in general require that the protocol be fully specified in advance. This does not include just the maximum number of allowed rescans NsN_s, but also the thresholds bib_i and the conditions uiu_i during rescans that were never performed. Incidentally, this is also true for the likelihood method that we are proposing here, though as we will see the effect of conservative assumptions on the ensemble of experiments is minor.

The Bayesian Power-Measured (BPM) framework proposed by Palken et al. [8] tackles this issue by means of a hybrid approach between Bayesian and frequentist inference. For each scan ii performed and power excess XiX_i collected, equation (9) of Ref. [8] defines:

Λi(A)≡P(Xi∣A)P(Xi∣A=0).\begin{equation} \Lambda_i(A)\equiv \frac{\mathbb{P}(X_i\mid A)}{\mathbb{P}(X_i\mid A=0)}. \tag*{(3)} \end{equation}

For sequential independent measurements, evidence collected through the scans accumulates multiplicatively through the total update Λ(A)=∏iΛi(A)\Lambda(A)=\prod_i\Lambda_i(A)—this, regardless of any specifications on NsN_s, bib_i, or uiu_i.

This quantity can be interpreted both in frequentist and Bayesian context; in particular, it can be viewed as both a likelihood ratio [10] and as a Bayes factor of AA against the null in the limit of infinitesimal priors. It can also provide evidence either for or against a specified signal strength in both Bayesian [e.g. 43, 44] and frequentist frameworks. In the latter, test and rejection of hypotheses on AA are based on a threshold condition on Λ(A)\Lambda(A). Ref. [8] connects the Bayesian interpretation to frequentist exclusion through the geometric (BT2) framework. The result is then extended to BPM, but without directly calibrating its false-exclusion probability.

In fact, in the BT2 framework ΛBT2\Lambda^{\scriptscriptstyle \mathrm{BT2}} can take only two values (success/fail). As equation (A8) of Ref. [8] shows, in a small false-positive regime ΛfailBT2\Lambda_{\rm fail}^{\scriptscriptstyle \mathrm{BT2}} approximates the total false negative rate FnF_{\rm n}. Hence, ΛfailBT2=0.1\Lambda_{\rm fail}^{\scriptscriptstyle \mathrm{BT2}}=0.1 approximately corresponds to 90%90\% exclusion. In BPM, Λ\Lambda becomes a ratio of continuous likelihoods. While it is argued that the interpretation of the update is unchanged, Λ≃Fn\Lambda\simeq F_{\rm n} relies on the binary space and does not necessarily extend to continuous power measurements. The normalization of Λ\Lambda provides a shared scale for the relative evidence across BT2 and BPM, but not for the error probability.

This becomes evident by realizing the connection of BPM and the formalism of e-variables [e.g. 2, 4], justified by the fact that E0[Λi(A)]=1\mathbb{E}_0[\Lambda_i(A)]=1. The realization of an e-variable (i.e., e-value) is safe against type-I and -II errors thanks to the Markov inequalities:

P0(Λi(A)≥1/α)≤αandPA(Λi(A)≤λ)≤λ,\begin{equation} \mathbb{P}_0\left(\Lambda_i(A)\ge1/\alpha\right)\le\alpha \quad\text{and}\quad \mathbb{P}_A\left(\Lambda_i(A)\le\lambda\right)\le\lambda, \tag*{(4)} \end{equation}

with α\alpha, λ∈(0,1]\lambda\in(0,1]. Relations (4) apply also to the ensemble Λ(A)\Lambda(A), provided each Λi(A)\Lambda_i(A) shares the same null, is computed on uncorrelated data, and satisfies E0[Λ(A)]≤1\mathbb{E}_0[\Lambda(A)]\le 1 conditioned on the result of the previous scan [3, 2].

As a consequence, Λ(A)≤λ\Lambda(A)\le\lambda guarantees a false-exclusion probability no larger than λ\lambda, but the bound is not necessarily saturated. For instance, a 1-scan measurement has a maximal false-exclusion probability Φ(−κ)\Phi(-\kappa) with κ=2ln⁡(1/λ)\kappa = \sqrt{2\ln(1/\lambda)}. This probability is only 1.6%1.6\% for λ=0.1\lambda=0.1, rather than 10%.

Figure 1. Coverage as a function of θ=Au1\theta=\mathbb{{A}}\sqrt{u_1} for the 1-scan BPM and the 2-scan likelihood “mix” frameworks. Calibrated value λ0.1\lambda_{0.1} for BPM comes from equation (5)—see text for details.

In general, λα\lambda_\alpha (or κα\kappa_\alpha) can be calibrated in a frequentist sense, such that

λα=sup⁡{λ>0:sup⁡A>0PA(Λ(A)≤λ)≤α}.\begin{equation} \lambda_\alpha = \sup\left\{\lambda>0: \sup_{\mathbb{A}>0}\mathbb{P}_\mathbb{A}(\Lambda(\mathbb{A})\le\lambda)\le\alpha \right\}. \tag*{(5)} \end{equation}

This prescribes a minimal coverage of (1−α)(1-\alpha) at the least-favorable A\mathbb{A} at the expenses of procedural freedom.

Likelihoood-based upper limits

If procedural freedom is to be sacrificed, a frequentist construction can provide an exact exclusion probability. The starting point of the framework presented here is the same as BPM—that is, a series of (re)scanned power excesses XiX_i provided e.g. by the application of a likelihood-ratio test [e.g. 7, 9] or sequential transformations on the power spectra [e.g. 5, 6].

Following Ref. [9], we define cumulative variables:

Sm=∑i=1mXi∑i=1mui=X[m]u[m]∼N(θm,1),\begin{equation} S_m = \frac{\sum_{i=1}^{m} X_i}{\sqrt{\sum_{i=1}^{m} u_i}}= \frac{X_{[m]}}{\sqrt{u_{[m]}}}\sim\mathcal{N}(\theta_m,1), \tag*{(6)} \end{equation}

so that the trigger condition Sm≥βmS_m\ge\beta_m automatically includes prior evidence from the (m−1)(m-1) previous scans:

Sm≥βm⟺X[m]≥βmu[m].\begin{equation} S_m\ge\beta_m \quad\Longleftrightarrow\quad X_{[m]}\ge \beta_m \sqrt{u_{[m]}}. \tag*{(7)} \end{equation}

The variables SmS_m are correlated according to

ρjk≡Cov(Sj,Sk)=u[min⁡(j,k)]u[max⁡(j,k)].\begin{equation} \rho_{jk} \equiv {\rm Cov}(S_j,S_k) = \sqrt{\frac{u_{[\min(j,k)]}}{u_{[\max(j,k)]}}}. \tag*{(8)} \end{equation}

Couplings θm=Au[m]\theta_m=\mathbb{A}\sqrt{u_{[m]}} and ϑm=Au[m]\vartheta_m=A\sqrt{u_{[m]}} are not independent parameters, but branch-dependent representations of A\mathbb{A} and AA. Hence, we define the dimensionless strengths θ≡θ1\theta\equiv\theta_1 and ϑ≡ϑ1\vartheta\equiv\vartheta_1, so that θm=θu[m]/u1\theta_m = \theta\sqrt{u_{[m]}/u_1}.

To test a value ϑ\vartheta given M=mM=m performed scans, an observed Sm=smS_m=s_m, and an unconstrained maximum-likelihood estimator (MLE) ϑ^m=sm\hat\vartheta_m=s_m, we use the one-sided likelihood-ratio statistic:

qϑ=qϑ(ϑ^m,m)={(ϑm−ϑ^m)2,ϑ^m≤ϑm;0,ϑ^m>ϑm.\begin{equation} q_\vartheta = q_\vartheta(\hat{\vartheta}_m, m) = \begin{cases} (\vartheta_m-\hat\vartheta_m)^2, & \hat\vartheta_m\leq\vartheta_m;\\ 0, & \hat\vartheta_m > \vartheta_m. \end{cases} \tag*{(9)} \end{equation}

The case in which ϑ\vartheta is constrained to be non-negative is described by the statistic qϑ+q_{\vartheta}^{+} defined in (13).

The pp-value is given by the sum over all the NsN_s mutually exclusive stopping stages mm:

pϑ=∑m=1NsPϑ(M=m ,qϑ(ϑ^,m)≥qϑobs),\begin{equation} p_\vartheta = \sum_{m=1}^{N_s} \mathbb{P}_\vartheta\left(M = m \,, q_\vartheta(\hat{\vartheta}, m) \geq q_{\vartheta}^{\mathrm{obs}}\right), \tag*{(10)} \end{equation}

where Pϑ\mathbb{P}_{\vartheta} denotes probabilities evaluated assuming the tested signal strength ϑ\vartheta, and qϑobsq_{\vartheta}^{\rm obs} is the statistic evaluated on the observed experimental outcome.

Since qϑq_{\vartheta} (and qϑ+q_{\vartheta}^+) are monotonically decreasing for ϑ^m≤ϑm\hat\vartheta_m\leq\vartheta_m, condition (10) translates as a single threshold on ϑ^m\hat\vartheta_m, whose expression is given in Appendix § 6.

For Ns=1N_s=1, equation (10) reduces to

pϑ(1)=Pϑ(S1≤s1)=Φ(s1−ϑ).\begin{equation} p_\vartheta^{(1)} = P_\vartheta(S_1\leq s_1) = \Phi(s_1-\vartheta). \tag*{(11)} \end{equation}

Imposing pϑ(1)≡αp_\vartheta^{(1)}\equiv\alpha gives the upper limit at (1−α)(1-\alpha) CL: \vartheta_{\text{\textsc{ul}}}^{\ell1} = s_1+\Phi^{-1}(1-\alpha) for the unconstrained case, and \vartheta_{\text{\textsc{ul}}}^{\ell 1+}=\max\!\left(0,\vartheta_{\rm UL}^{\ell 1}\right) for the constrained one.

For Ns=2N_s=2, it is convenient to define Bm(ϑ)≡βm−ϑmB_m(\vartheta)\equiv\beta_m-\vartheta_m and Rm(ϑ)≡rm(ϑ)−ϑmR_m(\vartheta)\equiv r_m(\vartheta)-\vartheta_m, with rm(ϑ)r_m(\vartheta) defined in (14). The pp-value is then given by the close form:

pϑ(2)=Φ ⁣[min⁡(B1,R1)]+Φ(R2)−Φ2(B1,R2;ρ12),\begin{equation} p_\vartheta^{(2)}=\Phi\!\left[\min(B_1,R_1)\right]+ \Phi(R_2)-\Phi_2(B_1,R_2;\rho_{12}), \tag*{(12)} \end{equation}

where Φ2\Phi_2 is the standard bivariate normal CDF with correlation ρ12\rho_{12} (8). Derivation of equation (12) and its extension to Ns>2N_s>2 are given in the Appendix.

The frequentist two-scan upper limit is defined implicitly by imposing equation (12) to be equal to α=1−CL\alpha = 1 - \rm CL. The same relation applies to the constrained case, provided that Rm+(ϑ)≡rm+(ϑ)−ϑmR_m^+(\vartheta)\equiv r_m^+(\vartheta)-\vartheta_m is used.

To characterize the behavior of this framework, we consider the expected upper limit, \mathbb{E}[\vartheta_{\text{\textsc{ul}}}], which is computed in Appendix § 8. We also define the dimensionless quantities β=b1/u1\beta = b_1/\sqrt{u_1} and ω=u2/u1\omega=u_2/u_1. The former defines the rescan trigger; the latter controls the amount of information from a possible second scan: ω→0\omega\to0 recovers the one-scan result (uninformative second scan) while ω→∞\omega\to\infty represent a second scan of arbitrary precision.

Figure 2. Top: Expected upper limit as a function of β\beta, for θ=0\theta = 0 and 90%90\% CL. Bottom: Relative difference between the limiting cases ω→0\omega\to0 and ω→∞\omega\to\infty (solid) and between the “mix” and ω→∞\omega\to\infty unconstrained likelihood approach (dashed). The geometric test reaches \mathbb{E}[\vartheta_{\text{\textsc{ul}}}] = 6.3 for β=5\beta = 5.

For Ns=2N_s = 2 and ω→∞\omega\to\infty, experiments that stop after the first scan incur a penalty from the unrealized branch with respect to the one-scan-only case, while realizations that trigger a rescan benefit from its arbitrary precision. Therefore, \mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\ell2}]<\mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\ell1}] if β\beta is low enough to favor rescans. If not, \mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\ell2}]>\mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\ell1}] and \mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\ell2}]\to\mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\ell1}] as β\beta increases, since rescans become increasingly rare and the probability assigned to the second branch vanishes. In general, in the regime of rare rescans, \mathbb{E}[\vartheta_{\text{\textsc{ul}}}] weakens for increasing NsN_s, similarly to the geometric construction.

Since haloscopes operate in this latter regime, we can define a practical recipe (labeled “mix”) to deal with the unknown unrealized uiu_i: if m<Nsm<N_s scans are observed, quote the upper limit for Ns=m+1N_s=m+1 and um+1→∞u_{m+1}\to\infty. Under the assumptions of Figure 2, the relative difference induced by ω\omega in the likelihood upper limits is no higher than 6%6\% for β≳2\beta\gtrsim2, with the “mix” recipe worsening the un/constrained frameworks by a further 3.6% (0.4%) for β≥2\beta\ge2 (β≥3\beta\ge3) at most.

The “mix” framework is conservative also in its coverage (see Appendix § 7). As Figure 1 shows for Ns=2N_s = 2, Cmix(2)(θ)C_{\rm mix}^{(2)}(\theta) deviates from the nominal CL until both 1-scan rejections thresholds R1R_1 exceed the stopping threshold B1B_1—see equation (27). Moreover, under constraint θ≥0\theta\ge 0, coverage is unity at the boundary θ=0\theta = 0. The transition is continuous, with changes in slope caused by these threshold crossings. For more than two scans, the coverage is determined by the sum of the branch contributions in equation (26), and the conservatism must be assessed for the chosen protocol.

Comparison among frameworks

In this section we extend the comparison to the other frameworks, while keeping Ns≤2N_s\le 2. Similar considerations hold for greater NsN_s.

In the geometric framework, the one-scan limit corresponds to ω→∞\omega\to\infty, when a nonzero signal would trigger a positive experimental outcome (success) with probability 1. In order not to introduce an additional second threshold b2b_2 just for this framework, we assign the same false-positive rate β\beta to both scans. This choice matches the spirit of Refs. [5, 6] and experimental practice, but it causes ω→0\omega\to 0 to be singular. A finite threshold cannot be overcome by a second scan carrying no information. However, this regime is of little experimental relevance.

For BPM, some care is needed in defining an equivalent two-scan comparison, since its e-value construction does not require the stopping rule to be fixed in advance. We therefore consider two implementations. The first one, labeled “fix,” follows the same stopping protocol as the likelihood-based construction, and the condition Λ(ϑ)≤λ\Lambda(\vartheta)\leq\lambda is evaluated only at the end, when the procedure halts. In the second one, labeled “any,” the same condition is checked after every acquisition, and a value is excluded as soon as the cut is crossed.

The expected upper limit \mathbb{E}[\vartheta_{\text{\textsc{ul}}}] for the different frameworks, as a function of β\beta, for θ=0\theta = 0 and 90%90\% CL is shown in Figure 2. The likelihood framework considers both the constrained and unconstrained case. The BPM upper limit, on the other hand, has been computed both for λ=1−CL\lambda = 1-\rm CL and with coverage calibrated according to (5). In the specified regime, ϑ≤ϑc≡β+β2+κ2\vartheta\le\vartheta_c\equiv\beta+\sqrt{\beta^2+\kappa^2} (see Appendix § 8) therefore BPM-fix = BPM-any ≡\equiv BPM. The BPM framework gives an upper limit which is ∼20%\sim 20\% higher than the likelihood one when the coverage is calibrated, up to ∼80%\sim 80\% higher when it is not.

Figure 3. Expected upper limits at 90%90\% CL on the ∣Caγ∣|C_{a\gamma}| coupling achievable by ALPHA [42] (see text). The line-widths/shaded regions represent the interval 3≤β≤53\le\beta\le 5.

In order to provide also a physical benchmark, Figure 3 shows the upper limit achievable by ALPHA (Axion Longitudinal Plasma HAloscope) targeting the 10–20 GHz (40–80 μeV) range while operating at the conditions discussed in Ref. [42]. The statistical assumptions are those adopted throughout this section, with ω→∞\omega\to\infty, and β\beta from 3 to 5. Here, the uncalibrated BPM (Geometric) upper limits are 30% (115%) higher (i.e. less constraining) compared to the Likelihood framework, while the application of the “mix” recipe would increase their limits by ∼0.5%\sim 0.5\%.

Conclusions

Statistical inference for axion haloscopes has one noteworthy feature: the possibility of sequentially acquiring new data in so-called rescans. In this work, we showed that a conventional likelihood approach yields substantially stronger expected limits.

The challenge of frequentist inference for axion haloscopes is specifying in advance all possible outcomes of the experiment, including stopping rules and detector configurations for scans that may never be performed. A fully Bayesian approach, defining the appropriate Bayes factor, is a viable solution. However, insisting on frequentist coverage may yield overly conservative results.

A fully frequentist approach is also a viable solution. We showed that, in experimental regimes, uncertainty about future detector configurations has only a limited impact, even when all the possible experimental outcomes are included. We also proposed a practical “mix” prescription to reduce dependence on unrealized detector configurations while keeping its statistical performance explicitly assessable.

We performed explicit quantitative comparisons for Ns≤2N_s\le2, while the likelihood construction and coverage expressions extend to Ns>2N_s>2 with the same conclusions. In regimes of rare rescans, the contribution by pp-values from higher scans becomes negligible by construction. Future work will address the problem of discovery.

Acknowledgments

The authors thank Jón Gumundsson, Junu Jeong, Andrew Meyer, and Michael Sekatchev for their feedback and comments. The authors acknowledge valuable discussions within the ALPHA Collaboration. The authors acknowledge support from the Knut and Alice Wallenberg Foundation and Olle Engkvists Foundation. AGR acknowledges support from the EDUCATE excellence centre funded by the Swedish Research Council through grant Dnr 2022-06627, and the use of Claude (Anthropic) and ChatGPT (OpenAI)—as a help to brainstorm and support the central idea and the theoretical framework, to assist in coding the numerical studies, and to edit the manuscript. The authors take full responsibility for the content of this work.

Likelihood-Ratio tests and p\bm p-values

In this section, we derive the likelihood-ratio statistics, the corresponding thresholds on the scan-dependent maximum-likelihood estimator and the explicit expression for the pp-values.

For an outcome consisting of M=mM=m scans, with an observed value Sm=smS_m = s_m, as per (6), the log-likelihood is: ℓm(ϑ)=−(sm−ϑm)2/2+const.\ell_m(\vartheta)=-(s_m-\vartheta_m)^2/2+{\rm const}.

The likelihood-ratio statistic qϑ=2[ℓ(ϑ^m)−ℓ(ϑm)]q_\vartheta=2[\ell(\hat\vartheta_m)-\ell(\vartheta_m)] has then the form given in (9). On the other hand, if the physical parameter is constrained to be non-negative, the MLE becomes ϑ~m=max⁡(0,ϑ^m)\tilde\vartheta_m= \max(0,\hat\vartheta_m), which defines

qϑ+={(ϑm2−2ϑmϑ^m),ϑ^m<0;(ϑm−ϑ^m)2,0≤ϑ^m≤ϑm;0,ϑ^m>ϑm.\begin{equation} q_\vartheta^+ = \begin{cases} (\vartheta_m^2 - 2\vartheta_m\hat\vartheta_m), & \hat\vartheta_m < 0;\\ (\vartheta_m - \hat\vartheta_m)^2, & 0 \leq \hat\vartheta_m\leq\vartheta_m;\\ 0, & \hat\vartheta_m > \vartheta_m. \end{cases} \tag*{(13)} \end{equation}

The condition qϑ≥qϑobsq_\vartheta\ge q_\vartheta^{\rm obs} defining the p-value (10) can be expressed as

ϑ^m≤r(ϑm)≡ϑm−qϑobs,\begin{equation} \hat\vartheta_m \leq r(\vartheta_m) \equiv\vartheta_m - \sqrt{q_\vartheta^{\mathrm{obs}}}, \tag*{(14)} \end{equation}

while the constrained case gives

ϑ^m≤r+(ϑm)≡{r(ϑm),qϑobs≤ϑm2;ϑm2−qϑobs2ϑm,qϑobs>ϑm2.\begin{equation} \hat\vartheta_m \leq r^+(\vartheta_m)\equiv \begin{cases} r(\vartheta_m), & q_\vartheta^{\rm obs}\le \vartheta_m^2;\\ \dfrac{\vartheta_m}{2}-\dfrac{q_\vartheta^{\rm obs}}{2\vartheta_m}, & q_\vartheta^{\rm obs}>\vartheta_m^2 . \end{cases} \tag*{(15)} \end{equation}

The closed form in equation (12) follows directly by separating the mutually exclusive stopping branches; that is, for Ns=2N_s=2: pϑ(2)=pϑ,1(2)+pϑ,2(2).p_\vartheta^{(2)}=p_{\vartheta,1}^{(2)}+p_{\vartheta,2}^{(2)}. Using the standardized quantities BmB_m, RmR_m, and Zm≡Sm−ϑmZ_m\equiv S_m -\vartheta_m, we get:

pϑ,1(2)=Pϑ(Z1<B1,Z1≤R1)=Φ ⁣[min⁡(B1,R1)].\begin{equation} p_{\vartheta,1}^{(2)} = \mathbb{P}_\vartheta \left(Z_1<B_1,Z_1\leq R_1 \right) = \Phi\!\left[\min(B_1,R_1)\right]. \tag*{(16)} \end{equation}

For the second branch, we have

pϑ,2(2)=Pϑ(Z1≥B1, Z2≤R2)=Pϑ(Z2≤R2)−Pϑ(Z1<B1,Z2≤R2)=Φ(R2)−Φ2(B1,R2;ρ12).\begin{align} p_{\vartheta,2}^{(2)} &=%\label{eq:p22_def} \mathbb{P}_\vartheta \left( Z_1\geq B_1,\, Z_2\leq R_2 \right) \tag*{(17)}\\ &= \mathbb{P}_\vartheta(Z_2\leq R_2) - \mathbb{P}_\vartheta (Z_1<B_1,Z_2\leq R_2) \tag*{(18)}\\ &= \Phi(R_2) - \Phi_2(B_1,R_2;\rho_{12}). \tag*{(19)} \end{align}

The same construction extends recursively to an arbitrary number of possible scans. For Ns=3N_s=3, the first contribution remains unchanged: pϑ,1(2)=pϑ,1(3)p_{\vartheta,1}^{(2)}=p_{\vartheta,1}^{(3)}. The second branch is reached only after the first threshold has been exceeded and the second threshold has not, while at the same time satisfying the likelihood-ratio ordering:

pϑ,2(3)=Pϑ(Z1≥B1, Z2≤min⁡(B2,R2))=Φ ⁣[min⁡(B2,R2)]−Φ2 ⁣[B1,min⁡(B2,R2);ρ12].\begin{align} p_{\vartheta,2}^{(3)} &= \mathbb{P}_\vartheta \left( Z_1\geq B_1,\, Z_2\leq\min(B_2,R_2) \right) \tag*{(20)}\\ &= \Phi\!\left[\min(B_2,R_2)\right] -\Phi_2\!\left[B_1,\min(B_2,R_2); \rho_{12}\right]. \tag*{(21)} \end{align}

Finally, reaching the third scan requires the first two thresholds to have been exceeded. This gives:

pϑ,3(3)=Pϑ(Z1≥B1, Z2≥B2, Z3≤R3)= Φ(R3)−Φ2(B1,R3;ρ13)−Φ2(B2,R3;ρ23)+Φ3 ⁣(B1,B2,R3;ρ123),\begin{align} p_{\vartheta,3}^{(3)} =& \mathbb{P}_\vartheta \left( Z_1\geq B_1,\, Z_2\geq B_2,\, Z_3\leq R_3\right) \tag*{(22)}\\ =&\,\Phi(R_3) - \Phi_2(B_1,R_3;\rho_{13}) -\Phi_2(B_2,R_3;\rho_{23})+ \Phi_3\!\left( B_1,B_2,R_3; \bm{\rho}_{123} \right), \tag*{(23)} \end{align}

where we applied inclusion–exclusion [e.g. 12, 11]; Φ3\Phi_3 is the trivariate standard-normal CDF and ρ123\bm{\rho}_{123} is the correlation matrix of elements [ρ123]ij=ρij[\bm{\rho}_{123}]_{ij}=\rho_{ij} as per (8).

More generally, the contribution from a branch terminating at scan mm requires the trigger from all the previous (m−1)(m-1) thresholds, while preserving the likelihood-ratio condition (10). Therefore, the contribution of that branch is a joint probability of the series (Z1,…,Zm)(Z_1,\ldots,Z_m), evaluated in terms of Gaussian CDFs, and using the inclusion-exclusion principle to account for the previous (m−1)(m-1) trigger conditions. The constrained construction (θ≥0\theta\ge0) follows identically after replacing RmR_m by Rm+R_m^+.

On the coverage

In this section, we apply the branch-related pp-values to the computation of the coverage of the “mix” framework.

Coverage is defined as C(\theta) = \mathbb{P}_\theta\!\left(\vartheta_{\text{\textsc{ul}}}^{\rm mix}\ge\theta\right). Let pθ,m(Ns)(q)=Pθ(M=m, qθ>q)p_{\theta,m}^{(N_s)}(q)=\mathbb{P}_\theta(M=m,\,q_\theta>q) be the mthm^{\rm th} branch contributions evaluated at a generic threshold qq, under the full frequentist protocol at NsN_s allowed scans.

When the accepted ϑ\vartheta are all those \vartheta\le\vartheta_{\text{\textsc{ul}}}^{\rm mix}, noncoverage at the true value θ\theta is equivalent to rejecting θ\theta. Within branch M=mM=m, rejection occurs for qθ>qcritq_\theta>q^{\rm crit}, or equivalently Zm<Rm(θ;qmcrit)Z_m<R_m(\theta;q_m^{\rm crit}). For m=Nsm=N_s, qmcrit(θ)q^{\rm crit}_{m}(\theta) coincides with qmfull(θ)q^{\rm full}_{m}(\theta) from the full protocol: ∑iNspθ,i(Ns)(qNsfull)=α\sum_i^{N_s} p_{\theta,i}^{(N_s)}(q_{N_s}^{\rm full})=\alpha. On the other hand, and for m<Nsm<N_s, qcrit(θ)q^{\rm crit}(\theta) satisfies

lim⁡um+1→∞∑j=1m+1pθ,j(m+1) ⁣(qmcrit)=α.\begin{equation} \lim_{u_{m+1}\to\infty}\sum_{j=1}^{m+1} p_{\theta,j}^{(m+1)} \!\left(q^{\rm crit}_{m}\right) = \alpha. \tag*{(24)} \end{equation}

Then, summation over the mutually exclusive stopping branches gives

Cmix(Ns)(θ)=1−∑m=1Nspθ,m(Ns)(qmcrit),\begin{equation} C_{\rm mix}^{(N_s)}(\theta)=1-\sum_{m=1}^{N_s} p_{\theta,m}^{(N_s)}(q^{\rm crit}_{m}), \tag*{(25)} \end{equation}

where pθ,m(m+1)(qmcrit)≡pθ,m(Ns)(qmcrit)p_{\theta,m}^{(m+1)}(q^{\rm crit}_{m})\equiv p_{\theta,m}^{(N_s)}(q^{\rm crit}_{m}) for m<Nsm<N_s, since both protocols have identical stopping conditions and cumulative statistics through scan mm.

We can exploit again the exact coverage of the full construction, which gives

Cmix(Ns)(θ)=1−α+ ⁣∑m=1Ns−1 ⁣[pθ,m(Ns)(qNsfull)−pθ,m(Ns)(qmcrit)].\begin{equation} C_{\rm mix}^{(N_s)}(\theta)=1-\alpha+ \!\sum_{m=1}^{N_s-1}\! \left[ p_{\theta,m}^{(N_s)}(q_{N_s}^{\rm full}) -p_{\theta,m}^{(N_s)}(q^{\rm crit}_{m}) \right]. \tag*{(26)} \end{equation}

The same reasoning apply to the constrained construction (θ≥0\theta\ge 0). At the physical boundary, Cmix(Ns)(0)=1C_{\rm mix}^{(N_s)}(0)=1. If the full calibration is conservative rather than exact, Cfull(Ns)(θ)C_{\rm full}^{(N_s)}(\theta) replaces 1−α1-\alpha.

For Ns=2N_s = 2, relation (26) becomes

Cmix(2)(θ)=1−α+Φ ⁣[min⁡ ⁣(B1(θ),R1(θ,q2full))]−Φ ⁣[min⁡ ⁣(B1(θ),R1(θ,q1crit))],\begin{equation} \begin{aligned} C_{\rm mix}^{(2)}(\theta) =& 1-\alpha+\Phi\!\left[\min\!\left(B_1(\theta),R_1(\theta, q_2^{\rm full})\right)\right]\\ &- \Phi\!\left[\min\!\left(B_1(\theta),R_1(\theta, q_1^{\rm crit})\right)\right], \end{aligned} \tag*{(27)} \end{equation}

where q1critq_1^{\rm crit} satisfies equation (24).

Expected upper limits

In this section, we provide explicit expressions for the expected upper limits \mathbb{E}[\vartheta_{\text{\textsc{ul}}}], for the frameworks considered in this work and Ns≤2N_s\le 2.

In the geometric construction \vartheta_{\text{\textsc{ul}}}^{g_{N_s}} does not depend on the magnitude of the observed data. Hence, the value provided by (2) coincides with the expected benchmark.

For the likelihood framework with Ns=1N_s = 1, we derived \vartheta_{\text{\textsc{ul}}}^{\ell1 }=s_1+k_{\alpha}, where we defined kα=Φ−1(1−α)k_{\alpha}=\Phi^{-1}(1-\alpha). Replacing s1s_1 with its expected value θ\theta immediately gives \mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\ell1}]. In the constrained case:

\begin{equation} \mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\ell 1+}] = \int_{-k_\alpha}^{\infty} (s_1 + k_\alpha)\,\phi(s_1-\theta)\,\mathrm{d}s_1, \tag*{(28)} \end{equation}

where ϕ(s1−θ)\phi(s_1-\theta) is the distribution of s1s_1 and ϕ\phi is the standard normal CDF. Integration of (28) gives

\begin{equation} \mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\ell1+}] = (\theta + k_\alpha) \, \Phi(\theta + k_\alpha) + \phi(\theta + k_\alpha). \tag*{(29)} \end{equation}

Two branches contribute to \mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\ell 2}] = E_1 + E_2. Similarly to (28), E1E_1 comes from the one-scan outcome:

\begin{equation} E_1 = \int_{-\infty}^\beta \vartheta_{\text{\textsc{ul}}}(s_1)\,\phi(s_1 - \theta)\,\mathrm{d}s_1, \tag*{(30)} \end{equation}

where \vartheta_{\text{\textsc{ul}}}(s_1) solves pϑ(2)=αp_\vartheta^{(2)}=\alpha, and has to be evaluated numerically from equation (12). In the constrained case, \vartheta_{\text{\textsc{ul}}}^{+}(s_1) is the upper endpoint of the region pϑm=1+(2)≥αp_{\vartheta^+_{m=1}}^{(2)}\ge\alpha:

\begin{equation} \vartheta_{\text{\textsc{ul}}}^{+}(s_1)=\sup\left\{\vartheta^+_{m=1}\ge0:p_{\vartheta^+_{m=1}}^{(2)}\ge\alpha\right\}. \tag*{(31)} \end{equation}

Inequality (31) handles the case of strongly negative realization, for which there might not be a positive root.

For the contribution from the second branch, we have

\begin{equation} E_2 =\int_{-\infty}^{\infty}\vartheta_{\text{\textsc{ul}}}(s_2)\, f(s_2)\,\mathbb{P}(S_1>\beta\mid S_2 = s_2)\,\mathrm{d}s_2, \tag*{(32)} \end{equation}

where \vartheta_{\text{\textsc{ul}}}(s_2) differs from \vartheta_{\text{\textsc{ul}}}(s_1) in that the unconstrained MLE is now ϑ^=ρ12s2≡ρs2\hat{\vartheta}=\rho_{12} s_2\equiv\rho s_2. The value ϑ(s2)\vartheta(s_2) (and its constrained counterpart) are once again obtained from numerically solving the condition over pϑ(2)p_\vartheta^{(2)}.

The distribution of s2s_2, f(s2)f(s_2), comes directly from (6):

f(s2)=N(θ1+ω,1)=ϕ(s2−θ/ρ).\begin{equation} f(s_2)=\mathcal{N}(\theta\sqrt{1+\omega},1)=\phi(s_2-\theta/\rho). \tag*{(33)} \end{equation}

In order to compute P(S1>β∣S2=s2)\mathbb{P}(S_1>\beta\mid S_2 = s_2) we need the distribution of S1∣S2=s2S_1\mid S_2 = s_2. We apply the standard formulae for a partitioned multivariate normal distribution to S=(S1,S2)\bm{S}=(S_1, S_2)—see e.g. [1 Sec. 2.5]. This gives:

S1 ∣ S2=s2∼N(μs1+ρ(s2−μs2),1−ρ2)=N(ρs2,1−ρ2),\begin{align} S_1\,|\,S_2 &= s_{2}\sim\mathcal{N}(\mu_{s_1}+\rho(s_{2}-\mu_{s_2}), 1-\rho^2) \tag*{(34)}\\ &= \mathcal{N}(\rho s_{2}, 1-\rho^2), \tag*{(35)} \end{align}

and therefore:

P(S1>β∣S2=s2)=1−Φ ⁣(β−ρs21−ρ2).\begin{align} \mathbb{P}(S_1>\beta\mid S_2 = s_2) &=1-\Phi\!\left(\frac{\beta-\rho s_2}{\sqrt{1-\rho^2}}\right). \tag*{(36)} \end{align}

The same line of reasoning applies to Ns>2N_s>2 and to \mathbb{E}[\vartheta_{\text{\textsc{ul}}}^{\rm mix}]. In the latter case, contribution E1E_1 (30) from the one-scan branch would just require \vartheta_{\text{\textsc{ul}}}(s_1) to be substituted with \vartheta_{\text{\textsc{ul}}}^{\omega\to\infty}(s_1).

Concerning the BPM framework, we combine equation (9) of Ref. [8] with the definition λ=exp⁡(−κ2/2)\lambda = \exp(-\kappa^2/2). This allows to express the exclusion condition Λ(ϑ)≤λ\Lambda(\vartheta)\leq\lambda on the standardized variable ZmZ_m as Zm≤RmΛZ_m\leq R_m^\Lambda, where

R1Λ=ϑ−2θ2−κ22ϑandR2Λ=ϑ−2θ2ρ−ρκ22ϑ.\begin{equation} R_1^\Lambda = \frac{\vartheta-2\theta}{2} -\frac{\kappa^2}{2\vartheta} \quad\text{and}\quad R_2^\Lambda = \frac{\vartheta-2\theta}{2\rho} -\frac{\rho\kappa^2}{2\vartheta}. \tag*{(37)} \end{equation}

For BPM-fix, exclusion occurs either after stopping at the first scan, {Z1(θ)<B1(θ),  Z1(θ)≤R1Λ}\{Z_1(\theta)<B_1(\theta),\;Z_1(\theta)\leq R_1^\Lambda\}, or after a triggered rescan, {Z1(θ)≥B1(θ),  Z2(θ)≤R2Λ}\{Z_1(\theta)\geq B_1(\theta),\;Z_2(\theta)\leq R_2^\Lambda\}. For BPM-any, instead, a tested value is excluded as soon as the BPM cut is crossed, that is, {Z1(θ)≤R1Λ}\{Z_1(\theta)\leq R_1^\Lambda\} or {Z1(θ)≥B1(θ),  Z2(θ)≤R2Λ}\{Z_1(\theta)\geq B_1(\theta),\;Z_2(\theta)\leq R_2^\Lambda\}. Therefore, the probabilities associated to those events are:

Eθfix(ϑ)=Φ ⁣[min⁡(B1,R1Λ)]+Φ(R2Λ)−Φ2(B1,R2Λ;ρ),Eθany(ϑ)=Φ(R1Λ)+Φ(R2Λ)−Φ2 ⁣[max⁡(R1Λ,B1),R2Λ;ρ].\begin{align} E_\theta^{\rm fix}(\vartheta) &=\Phi\!\left[\min(B_1,R_1^\Lambda)\right] +\Phi(R_2^\Lambda) -\Phi_2(B_1,R_2^\Lambda;\rho),\nonumber\\ E_\theta^{\rm any}(\vartheta) &=\Phi(R_1^\Lambda)+\Phi(R_2^\Lambda) -\Phi_2\!\left[\max(R_1^\Lambda,B_1), R_2^\Lambda;\rho\right].\nonumber \end{align}

Finally, since E_\theta^k(\vartheta)=P_\theta(\vartheta_{\text{\textsc{ul}}}^k\le\vartheta), its complement is the survival function of the upper limit. Hence:

\begin{equation} \mathbb E_\theta[\vartheta_{\text{\textsc{ul}}}^k]= \int_0^\infty \left[1-E_\theta^k(\vartheta)\right]\,\mathrm{d}\vartheta, \tag*{(38)} \end{equation}

with k=k= fix, any. The two implementations differ only when the first scan already satisfies Λ(ϑ)≤λ\Lambda(\vartheta)\leq\lambda, triggers a rescan and then the second scan reverses that exclusion. The probability of this happening is

Pθ ⁣(β<S1≤ϑ2−κ22ϑ, S2>ϑ2ρ−ρκ22ϑ),\begin{equation} \mathbb{P}_\theta\!\left(\beta < S_1\le \frac{\vartheta}{2} -\frac{\kappa^2}{2\vartheta},\, S_2 > \frac{\vartheta}{2\rho} -\frac{\rho\kappa^2}{2\vartheta} \right), \tag*{(39)} \end{equation}

and the two implementations coincides for 2β>ϑ−κ2/ϑ2\beta> \vartheta -\kappa^2/\vartheta.

References

  1. [1]Anderson, T. W. (2003). An Introduction to Multivariate Statistical Analysis. John Wiley & Sons. https://doi.org/10.1002/0471722065
  2. [2]Grünwald, P., de Heide, R., & Koolen, W. (2024). Safe testing. Journal of the Royal Statistical Society Series B: Statistical Methodology, 86(5), 1091–1128. https://doi.org/10.1093/jrsssb/qkae011
  3. [3]Ville, J. (1939). Étude critique de la notion de collectif (Vol. 3). Gauthier-Villars. http://eudml.org/doc/192893eudml.org/doc/192893
  4. [4]Ramdas, A., & Wang, R. (2025). Hypothesis Testing with E-values. Foundations and Trends in Statistics, 1(1-2), 1–390. https://doi.org/10.1561/3600000002
  5. [5]Brubaker, B. M. (2017). First results from the HAYSTAC axion search. In Yale U.. https://arxiv.org/abs/1801.00835
  6. [6]Brubaker, B. M., Zhong, L., Lamoreaux, S. K., Lehnert, K. W., & van Bibber, K. A. (2017). HAYSTAC axion search analysis procedure. Phys. Rev. D, 96(12), 123008. https://doi.org/10.1103/PhysRevD.96.123008
  7. [7]Foster, J. W., Rodd, N. L., & Safdi, B. R. (2018). Revealing the Dark Matter Halo with Axion Direct Detection. Phys. Rev. D, 97(12), 123006. https://doi.org/10.1103/PhysRevD.97.123006
  8. [8]Palken, D. A., et al. (2020). Improved analysis framework for axion dark matter searches. Phys. Rev. D, 101(12), 123011. https://doi.org/10.1103/PhysRevD.101.123011
  9. [9]Gallo Rosso, A., Algeri, S., & Conrad, J. (2023). Sequential hypothesis testing for axion haloscopes. Phys. Rev. D, 108(2), 023003. https://doi.org/10.1103/PhysRevD.108.023003
  10. [10]Neyman, J., & Pearson, E. S. (1933). On the problems of the most efficient tests of statistical hypotheses. Philosophical Transactions of the Royal Society of London, 231A, 289–338. https://doi.org/10.1098/rsta.1933.0009
  11. [11]Genz, A., & Bretz, F. (2009). Computation of Multivariate Normal and $t$ Probabilities (Vol. 195). Springer-Verlag. https://doi.org/10.1007/978-3-642-01689-9
  12. [12]Feller, W. (1968). An Introduction to Probability Theory and Its Applications (Vol. 1). John Wiley & Sons.
  13. [13]Ahn, S., Lee, S., Choi, J., Ko, B. R., & Semertzidis, Y. K. (2021). Improved axion haloscope search analysis. JHEP, 04, 297. https://doi.org/10.1007/JHEP04(2021)297
  14. [14]Knirck, S., Millar, A. J., O'Hare, C. A. J., Redondo, J., & Steffen, F. D. (2018). Directional axion detection. JCAP, 11, 051. https://doi.org/10.1088/1475-7516/2018/11/051
  15. [15]Irastorza, I. G., & Redondo, J. (2018). New experimental approaches in the search for axion-like particles. Prog.\ Part.\ Nucl.\ Phys., 102, 89–159. https://doi.org/10.1016/j.ppnp.2018.05.003
  16. [16]Billard, J., et al. (2022). Direct detection of dark matter -- APPEC committee report. J.\ Phys.\ G, 49(4), 040501. https://doi.org/10.1088/1361-6471/ac5754
  17. [17]Irastorza, I. G. (2021). Laboratory searches for galactic axions. Rept.\ Prog.\ Phys., 84(10), 106901. https://doi.org/10.1088/1361-6633/ac0d1c
  18. [18]Semertzidis, Y. K., & Youn, S. (2019). Axion dark matter: How to see it?. SciPost Phys.\ Proc., 2, 021. https://doi.org/10.21468/SciPostPhysProc.2.021
  19. [19]Berlin, A., & Kahn, Y. (2025). New Technologies for Axion and Dark Photon Searches. Ann. Rev. Nucl. Part. Sci., 75(1), 83–108. https://doi.org/10.1146/annurev-nucl-121423-101015
  20. [20]Axion Dark Matter (Vol. 1). (2021). Axion Dark Matter (Vol. 1). https://doi.org/10.17181/CERN.97QY-CKH9
  21. [21]The Physics of Axions. (2022). The Physics of Axions. Springer. https://doi.org/10.1007/978-3-030-95852-7
  22. [22]Preskill, J., Wise, M. B., & Wilczek, F. (1983). Cosmology of the Invisible Axion. Phys. Lett. B, 120, 127–132. https://doi.org/10.1016/0370-2693(83)90637-8
  23. [23]Abbott, L. F., & Sikivie, P. (1983). A Cosmological Bound on the Invisible Axion. Phys. Lett. B, 120, 133–136. https://doi.org/10.1016/0370-2693(83)90638-X
  24. [24]Dine, M., & Fischler, W. (1983). The Not So Harmless Axion. Phys. Lett. B, 120, 137–141. https://doi.org/10.1016/0370-2693(83)90639-1
  25. [25]Green, D., et al. (2022). Snowmass Theory Frontier: Astrophysics and Cosmology. https://arxiv.org/abs/2209.06854
  26. [26]Arbey, A., & Mahmoudi, F. (2021). Dark matter and the early Universe: a review. Prog. Part. Nucl. Phys., 119, 103865. https://doi.org/10.1016/j.ppnp.2021.103865
  27. [27]Profumo, S., Giani, L., & Piattella, O. F. (2019). An Introduction to Particle Dark Matter. Universe, 5(10), 213. https://doi.org/10.3390/universe5100213
  28. [28]Feng, J. L. (2010). Dark Matter Candidates from Particle Physics and Methods of Detection. Ann. Rev. Astron. Astrophys., 48, 495–545. https://doi.org/10.1146/annurev-astro-082708-101659
  29. [29]Jaeckel, J., & Ringwald, A. (2010). The Low-Energy Frontier of Particle Physics. Ann. Rev. Nucl. Part. Sci., 60, 405–437. https://doi.org/10.1146/annurev.nucl.012809.104433
  30. [30]Bertone, G., Hooper, D., & Silk, J. (2005). Particle dark matter: Evidence, candidates and constraints. Phys. Rept., 405, 279–390. https://doi.org/10.1016/j.physrep.2004.08.031
  31. [31]Bergström, L. (2000). Nonbaryonic dark matter: Observational evidence and detection methods. Rept. Prog. Phys., 63, 793. https://doi.org/10.1088/0034-4885/63/5/2r3
  32. [32]Zhitnitsky, A. R. (1980). On Possible Suppression of the Axion Hadron Interactions. (In Russian). Sov.\ J.\ Nucl.\ Phys., 31, 260.
  33. [33]Dine, M., Fischler, W., & Srednicki, M. (1981). A Simple Solution to the Strong CP Problem with a Harmless Axion. Phys.\ Lett.\ B, 104, 199–202. https://doi.org/10.1016/0370-2693(81)90590-6
  34. [34]Shifman, M. A., Vainshtein, A. I., & Zakharov, V. I. (1978). Can Confinement Ensure Natural CP Invariance of Strong Interactions?. Phys.\ Lett.\ B, 78, 443–446. https://doi.org/10.1016/0370-2693(80)90209-6
  35. [35]Kim, J. E. (1979). Weak Interaction Singlet and Strong CP Invariance. Phys.\ Rev.\ Lett., 43, 103. https://doi.org/10.1103/PhysRevLett.43.103
  36. [36]Wilczek, F. (1978). Problem of Strong P and T Invariance in the Presence of Instantons. Phys.\ Rev.\ Lett., 40, 279–282. https://doi.org/10.1103/PhysRevLett.40.279
  37. [37]Weinberg, S. (1978). A New Light Boson?. Phys.\ Rev.\ Lett., 40, 223–226. https://doi.org/10.1103/PhysRevLett.40.223
  38. [38]Peccei, R. D., & Quinn, H. R. (1977). Constraints Imposed by CP Conservation in the Presence of Instantons. Phys.\ Rev.\ D, 16, 1791–1797. https://doi.org/10.1103/PhysRevD.16.1791
  39. [39]Peccei, R. D., & Quinn, H. R. (1977). CP Conservation in the Presence of Instantons. Phys.\ Rev.\ Lett., 38, 1440–1443. https://doi.org/10.1103/PhysRevLett.38.1440
  40. [40]Sikivie, P. (1983). Experimental Tests of the Invisible Axion. Phys. Rev. Lett., 51, 1415–1417. https://doi.org/10.1103/PhysRevLett.51.1415
  41. [41]Sikivie, P. (1985). Detection Rates for 'Invisible' Axion Searches. Phys. Rev. D, 32, 2988. https://doi.org/10.1103/PhysRevD.36.974
  42. [42]Bai, X., et al. (2026). Design of ALPHA Phase I: A Plasma Haloscope for 10--20 GHz Post-Inflation Axions. https://arxiv.org/abs/2608.10292
  43. [43]Jeffreys, H. (1998). The Theory of Probability. Oxford University Press.
  44. [44]Kass, R. E., & Raftery, A. E. (1995). Bayes Factors. Journal of the American Statistical Association, 90(430), 773–795. https://doi.org/10.1080/01621459.1995.10476572

Paper details

vlnc:W_fbjjk29w
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Claude, ChatGPT

View statement
Looking for related papers…
Contents