Quantum measurements lie at the core of quantum mechanics and are a cornerstone of interpretations of quantum theory1. Moreover, they have a crucial role in the evolution of quantum systems2 and have found appealing applications such as estimating unknown physical parameters using quantum resources3,4. In quantum information science, measurements are especially relevant for implementing quantum computing and communication protocols5. The simplest scenario involves two parties, where a sender can prepare and send information encoded in quantum states6. To access to this information, the receiver has to choose which quantum measurement will be performed7,8. Usually, the chosen measurement will depend on the properties of the received states and the features of the quantum protocol to be implemented. One possible task is identifying an unknown quantum state. One way to address this situation lies in the quantum tomography technique, which reconstructs the quantum state of an unknown physical system from the measured probabilities of a suitable set of observables. Nevertheless, the number of measurements required for a successful state reconstruction scales at least polynomially in the dimension of the state9, and requires multiple identical copies. Other tasks require identifying quantum states among others belonging to a given set in a single-shot measurement. However, performing this assignment is impossible deterministically when non-orthogonal states are considered.

Hence, quantum state discrimination (QSD) relies on identifying a quantum state belonging to a set of N known non-orthogonal quantum states. This task plays a fundamental role in several remarkable quantum protocols such as quantum key distribution10, quantum teleportation11,12, entanglement swapping13,14, and entanglement concentration15,16. Moreover, being a fundamental protocol, QSD also has been studied in relation with: contextuality17, path distinguishability18,19,20, and quantum correlations21,22,23. Thus, there are well-known strategies to implement QSD, namely the minimum error discrimination (MED)24,25, the unambiguous discrimination (UD)26,27,28,29, and the maximum confidence discrimination (MCD)30,31,32, each focused on optimizing some figure of merit33. To implement any QSD strategy, it is necessary first to determine the corresponding measurement defined by a set of positive operator-valued measures (POVMs)34,35, and then engineer a way to implement these POVMs in an experimental realization36,37. In the case of MED, there are N POVM elements \(\Pi _i\) associated with the discrimination of one of the states \(\rho _i\)7. However, since the given states are non-orthogonal, the discrimination process will inevitably introduce errors in identifying each quantum state \(\rho _i\)24. In the MED strategy, the average error probability is minimized25,38. Conversely, UD has no error in identifying each quantum state \(\rho _i\). This process can be achieved by introducing an additional POVM element, \(\Pi _0\), which is associated with an inconclusive outcome28. Finally, the MCD strategy maximizes the confidence in taking the measurement outcome i to indicate that the \(\rho _i\) state was prepared8. QSD has been experimentally implemented for MED39,40,41,42,43,44, UD45,46,47,48 and also for MCD49,50. Moreover, the experimental realization for maximizing the mutual information51 between two users was also performed52.

The discrimination of unknown quantum states also has been studied. In this case, the communicating parties have no information about the states to be discriminated. Remarkably, it has been demonstrated that discrimination is feasible even in this situation. This is achieved through a programmable discriminator53,54,55,56,57,58, that is, a device where the unknown states to be discriminated are stored. In particular, it has been shown that this programmable discriminator is universal and attains a success probability close to the optimum of unambiguous state discrimination. In addition, the discrimination between two known families of non-orthogonal quantum states has been studied from the point of view of neural networks59,60. Nevertheless, implementing it is highly costly since the device requires a priori stored information and additional ancilla systems to work properly.

Under certain conditions, the MED and UD strategies coincide with the MCD strategy31,61. Moreover, MED and UD can be joined simultaneously in a more general QSD scheme known as fixed rate of inconclusive outcomes (FRIO)62,63,64. In the case of QSD by FRIO, the average error probability in identifying the quantum state \(\rho _i\) is minimized under the condition of fixing the probability of inconclusive results65. Although for MED, UD and MCD there is no analytical solution to the discrimination by FRIO of N arbitrary non-orthogonal states66, in the case of two pure non-orthogonal states with arbitrary a priori preparation probabilities, the complete optimal solution is known61,67. Moreover, experimental schemes for the realization of FRIO onto two pure non-orthogonal states have been proposed68,69.

Here we present an experimental realization of FRIO discrimination on two pure non-orthogonal states encoded on the polarization state of single photons created by the spontaneous parametric down-conversion (SPDC) process. We use a double-path Sagnac interferometer to implement a genuine three-outcome POVM over a single polarization qubit performing the required measurement. A notable feature of our experimental setup is that it can be easily configured to implement the optimal FRIO POVM for all pairs of non-orthogonal states, arbitrary a priori probabilities, and any value of the fixed rate of the inconclusive outcome. In particular, it allows us to implement MED, UD, and any intermediate case of FRIO by fixing the rate of inconclusive outcomes, showing a good agreement with the theoretical results61,67. This experimental platform gives a flexible and customizable toolbox for a wide range of QSD strategies involving the FRIO scheme, which can be helpful in exploring novel applications for quantum information protocols and further research in quantum foundations.

Methods

We consider a qubit source of two pure non-orthogonal states denoted by \(|\phi _1\rangle\) and \(|\phi _2\rangle\), with a priori preparation probabilities \(\eta _1\) and \(\eta _2\), respectively. Without loss of generality, these states can be written as

$$\begin{aligned} |\phi _1\rangle&=\cos \alpha |0\rangle +\sin \alpha |1\rangle ,\\ |\phi _2\rangle&=\cos \alpha |0\rangle -\sin \alpha |1\rangle , \end{aligned}$$

where the states \(\{|0\rangle , |1\rangle \}\) represent the logical basis. The overlap between the states is given by the parameter \(s=\langle \phi _1|\phi _2\rangle =\cos (2\alpha )\), considering \(s\in [0,1]\). Moreover, the a priori probabilities can be set arbitrarily and they must satisfy the constraint \(\eta _1+\eta _2=1\). The FRIO discrimination process is carried out by using three POVM elements, where \(\Pi _{1(2)}\) is associated to the identification of \(|\phi _{1(2)}\rangle\) and \(\Pi _0\) corresponds to the inconclusive outcome, where no information of the states can be learned from the measurement. These three operators satisfy the condition

$$\begin{aligned} \Pi _{1}+\Pi _{2}+\Pi _{0}=\mathbbm {1}, \end{aligned}$$

where \(\mathbbm {1}\) is the identity operator. We define the following probabilities \(p_{1(2)}\), \(r_{1(2)}\) and \(q_{1(2)}\), corresponding to the probabilities of success, error and the inconclusive outcome, respectively, in the discrimination of \(|\phi _{1(2)}\rangle\). These probabilities encompass all possible outcomes in this case, that is \(p_{1(2)}+r_{1(2)}+q_{1(2)}=1\).

The average probabilities of success \(P_s\), error \(P_e\) and inconclusive outcome Q over the states, are given by61

$$\begin{aligned} \begin{aligned} P_s&=tr(\eta _1\rho _1\Pi _1)+tr(\eta _2\rho _2\Pi _2)=\eta _1p_{1}+\eta _2p_{2},\\ P_e&=tr(\eta _1\rho _1\Pi _2)+tr(\eta _2\rho _2\Pi _1)=\eta _1r_1+\eta _2r_2,\\ Q&=tr(\rho \Pi _0)=\eta _1q_1+\eta _2q_2, \end{aligned} \end{aligned}$$
(1)

where \(\rho _{1(2)}=|\phi _{1(2)}\rangle \langle \phi _{1(2)}|\), and \(\rho =\eta _1\rho _1+\eta _2\rho _2\). It is straightforward to see that the average probabilities satisfy \(P_s+P_e+Q=1\).

Optimal strategy for FRIO discrimination

The optimal FRIO strategy minimizes the average error probability \(P_e\) under the constraint that the inconclusive outcome probability Q is fixed. For this case, the optimal probabilities for FRIO were obtained by Bagan et al.61. Due to the symmetry of this task, it is enough to consider the case \(\eta _1\le \eta _2\), that is, when \(0\le \eta _1\le 1/2\). The FRIO solution identifies three intervals where the optimal probabilities can be obtained. These intervals depend on the overlap values s and the probability \(\eta _1\) given a fixed value of Q61,67. Interval I is defined as

$$\begin{aligned} \frac{s^2}{1+s^2}\le \eta _1\le 1/2\quad \text {and}\quad 0\le Q\le Q_0, \end{aligned}$$

where \(Q_0=2s\sqrt{\eta _1\eta _2}\) is the maximum inconclusive event probability for this interval. Interval II is defined as

$$\begin{aligned} 0\le \eta _1\le \frac{s^2}{1+s^2}\text {and} 0\le Q\le Q_{th}, \end{aligned}$$

with \(Q_{th}=\frac{2\eta _1\eta _2(1-s^2)}{1-Q_0}\) a threshold value that separates intervals II and III. The optimal probabilities for these intervals are given by

$$\begin{aligned} q_i&=\frac{Q}{2\eta _i},\\ r_i&=\frac{1}{2}\left( 1-q_i-\frac{(1-q_i){\overline{Q}}-\frac{(Q_0-Q)^2}{2\eta _i}}{\sqrt{{\overline{Q}}^2-(Q_0-Q)^2}}\right) ,\\ p_i&=1-q_i-r_i, \end{aligned}$$

for \(i=1,2,\) and \({\overline{Q}}=1-Q\). Therefore, the optimal error probability \(P_e\) valid in intervals I and II is minimal and reads

$$\begin{aligned} P_e=\frac{1}{2}\left( {\overline{Q}}-\sqrt{{\overline{Q}}^2-(Q_0-Q)^2}\right) . \end{aligned}$$

On the other hand, the interval III is defined when

$$\begin{aligned} 0\le \eta _1\le \frac{s^2}{1+s^2}\quad \text {and}\quad Q_\mathrm{th}\le Q\le \eta _1+\eta _2s^2, \end{aligned}$$

and the optimal probabilities are given by

$$\begin{aligned} p_1= & {} 0, r_1=\frac{P_e}{\eta _1}, q_1=1-r_1,r_2=0,\\ p_2= & {} \left( s\sqrt{r_1}+\sqrt{(1-r_1)(1-s^2)}\right) ^2, q_2=1-p_2, \end{aligned}$$

and the optimal average error probability is given by67

$$\begin{aligned} P_e=\frac{1}{1-4c}\left( \eta _1{\overline{Q}}+c(\eta _2-\eta _1-2{\overline{Q}})-Q_0\sqrt{c(Q{\overline{Q}}-c)}\right) , \end{aligned}$$

where \(c=\eta _1\eta _2(1-s^2)\). Note that the optimal strategy in interval III is implemented using a two-outcome projective measurement.

A notable feature of FRIO discrimination is that allows not only MED (when \(Q=0\)) and UD (when \(Q=Q_{max}\)), but also intermediate cases. Here, \(Q_{max}\) is the maximum possible value for Q in each interval defined above. To show the versatility of this state discrimination scheme, we present an optical experiment performing FRIO for both cases (MED and UD), and we consider an intermediate case for the inconclusive event \(Q=Q_0/2\) using the same device implementing the required POVMs in Eq. (1) for each interval. The optimal POVM implementation depends on the overlap s between the non-orthogonal qubit states \(\{|\phi _i\rangle \}\). These states are prepared by encoding the polarization modes (horizontal and vertical directions) of single photons generated using a heralded source based on the SPDC process. We are able to finely tuning the parameter s and the a priori preparation probabilities \(\eta _i\), performing FRIO strategies (\(Q=0\), \(Q=Q_0/2\), and \(Q=Q_{max}\)) considering seven values of the overlap s in two preparation scenarios, when \(\eta _1=\eta _2\) and \(\eta _1<\eta _2\).

Experimental description

Figure 1
figure 1

Experimental setup used to implement FRIO quantum state discrimination. Alice can prepare the non-orthogonal state \(\left| \phi _i\right\rangle\) using the half-wave plate \(HWP(\theta )\), encoding on a single photon generated at the heralded source based on the SPDC process. Bob uses a two-path Sagnac interferometer to perform the three-outcomes POVM needed for the chosen discrimination strategy covered by the FRIO scheme. Each outcome is associated with success, error, and inconclusive results. In particular, for intervals I and II, the inconclusive outcome Q is detected in the spatial mode labeled by \(\left| V2\right\rangle\). Thus, the state \(\left| \phi _{1}\right\rangle\) is identified when \(\left| V1\right\rangle\) is detected, while the \(\left| \phi _{2}\right\rangle\) is associated to detecting in \(\left| H1\right\rangle\). See the main text for more details.

Full size image

To carry out FRIO quantum state discrimination, we implement a heralded photon source based on the spontaneous parametric down-conversion (SPDC) process and a two-path Sagnac interferometer to perform the required measurement. The experimental setup is depicted in Fig. 1. A continuous-wave laser at 405 nm pumps a type-II nonlinear periodically poled potassium titanyl phosphate (PPKTP) crystal to create degenerate down-converted photons at 810 nm with horizontal and vertical polarization. To ensure the degenerate phase-matching conditions and remove the remaining pump beam, a Semrock high-quality narrow bandpass filter centered at 810 nm is used, with 0.5 nm of bandwidth and a peak transmission of \(> 90\%\). To maximize the coincidence count rate, we consider a numerical model70. Precisely, the optimal coupling condition is reached when \(\omega _{DCP}=\sqrt{2}\omega _{p}\), where \(\omega _{p}\) and \(\omega _{DCP}\) are the waist modes of the pump beam and the down-converted photons at the center of the PPKTP crystal, respectively. In our case, these waists are adjusted using a 20 cm focal length lens L and \(10\times\) objective lenses.

The generated photon pairs can be used to implement a heralded single-photon source, in the sense that one down-converted photon arrives at the trigger detector announcing the passage of the other photon through the stages of Alice and Bob (see Fig. 1). Therefore, a polarizing beam splitter (PBS) is placed after the PPKTP crystal to separate the down-converted photons deterministically. Then, they are sent to the trigger and Alice section coupling into single-mode optical fibers (SMF), removing any spatial correlation between them that could arise from imperfections when satisfying the phase-matching conditions in the crystal. Moreover, to maintain the polarization state of the photons through the propagation in the optical fiber, we use a manual fiber polarization controller for the photon arriving at Alice, and polarizing films are placed in front of the trigger and Alice’s detectors to ensure the correct polarization mode of the detected state.

Alice can prepare the two non-orthogonal states \(\{\left| \phi _1\right\rangle ,\left| \phi _2\right\rangle \}\) using the half-wave plate HWP(\(\theta\)) to encode a polarization qubit her photon. The polarization qubit states read

$$\begin{aligned} \begin{aligned} |\phi _1\rangle&=\cos 2\theta |H\rangle +\sin 2\theta |V\rangle ,\\ |\phi _2\rangle&=\cos 2\theta |H\rangle -\sin 2\theta |V\rangle , \end{aligned} \end{aligned}$$
(2)

where \(|H\rangle\) and \(|V\rangle\) are the horizontal and vertical polarization modes, and \(\theta\) is the inclination angle of the HWP with respect to its fast axis. The photons are sent to Bob through free space to implement the FRIO state discrimination procedure. To generate the global unitary transformation U over the non-orthogonal states required to implement the POVM69, we resort to couple the polarization degree of freedom with two spatial propagation modes as an ancilla system. First, Bob rotates the polarization state using a half-wave plate \(\hbox {HWP}_{{1}}\) oriented at angle \(\theta _{1}\) and then inputs the state to the two-path Sagnac interferometer configuration, which is composed of three laser mirrors, the \(\hbox {HWP}_{{2}}\) and a PBS. In this device the polarization is coherently coupled with the spatial modes71,72,73, since the PBS splits the incident photon through the clockwise (reflected) or counterclockwise (transmitted) mode inside the interferometer. Thus, the PBS operation can be seen as a controlled-NOT gate: the photon populates a spatial mode depending on the input polarization state. Moreover, in the counterclockwise path, \(\hbox {HWP}_{{2}}\) is aligned at angle \(\theta _{2}\), which rotates the photon’s polarization again if it propagates in this spatial mode. Then, a new passage through the same PBS superposes the two spatial modes. We denote the initial and second photon propagation paths by \(|1\rangle\) and \(|2\rangle\). Lastly, an \(\hbox {HWP}_{{3}}\) at \(\theta _{3}\) and a PBS are placed in the \(|1\rangle\) mode to obtain three outcomes that determine the required POVM. The global unitary can be written as

$$\begin{aligned} U=C(\theta _{3}) \cdot CNOT \cdot C(\theta _{2}) \cdot CNOT \cdot C(\theta _{1}), \end{aligned}$$

where the transformation \(C(\theta _{i})\) represents a rotation of the polarization in an angle \(2\theta _{i}\). Then, applying U onto the non-orthogonal states \(\{|\phi _1\rangle ,|\phi _2\rangle \}\) gives us the following transformation,

$$\begin{aligned} \begin{aligned} U|\phi _1\rangle |1\rangle&=\sqrt{p_1}|V\rangle |1\rangle +\sqrt{r_1}|H\rangle |1\rangle +\sqrt{q_1}|V\rangle |2\rangle ,\\ U|\phi _2\rangle |1\rangle&=\sqrt{r_2}|V\rangle |1\rangle +\sqrt{p_2}|H\rangle |1\rangle +\sqrt{q_2}|V\rangle |2\rangle , \end{aligned} \end{aligned}$$
(3)

where the parameters \(p_i\), \(r_i\), and \(q_i\) are the optimal success, error, and inconclusive outcome probabilities associated to discriminate \(|\phi _i\rangle\). Indeed, for the case when \(\eta _{1}=\eta _{2}=1/2\) (that is, equal state preparation), we can write these optimal probabilities in terms of the waveplate angles \(\theta\) and \(\theta _1\), \(\theta _2\), \(\theta _3\):

$$\begin{aligned} \begin{aligned} p_i&= \frac{1}{2}(\cos 2\theta \cos \theta _{2}+\sin 2\theta )^{2},\\ r_i&= \frac{1}{2}(\cos 2\theta \cos \theta _{2}-\sin 2\theta )^{2},\\ q_i&= (\sin \theta _{2}\cos 2\theta )^{2}, \end{aligned}\end{aligned}$$
(4)

for \(i=1,2\), while \(\theta _{1}\) and \(\theta _{3}\) are fixed at 0 and \(\pi /4\), respectively. Remarkably, for these values of the initial state preparation probabilities \(\eta\), the optimal probabilities for FRIO discrimination correspond to the interval I for any inner product s between the states \(|\phi _i\rangle\).

On the other hand, for \(\eta _1<\eta _2\) we have that the optimal probabilities are given by

$$\begin{aligned} \begin{aligned} p_{1},r_{2}&= (\cos 2\theta (\sin \theta _{3}\cos \theta _{2}\cos \theta _{1}-\cos \theta _{3}\sin \theta _{1})\pm \sin 2 \theta (\sin \theta _{3}\cos \theta _{2}\sin \theta _{1}+\cos \theta _{3}\cos \theta _{1}))^{2},\\ p_{2},r_{1}&= (\cos 2\theta (\cos \theta _{3}\cos \theta _{2}\cos \theta _{1}+\sin \theta _{3}\sin \theta _{1})\pm \sin 2 \theta (\sin \theta _{3}\cos \theta _{1}-\cos \theta _{3}\cos \theta _{2}\sin \theta _{1}))^{2},\\ q_{1,2}&= \sin \theta _{2}(\cos 2\theta \cos \theta _{1}\pm \sin 2\theta \sin \theta _{1})^{2}, \end{aligned} \end{aligned}$$
(5)

where the sign ± is taken according to the state labeled by \(i=1,2\) to be discriminated in the FRIO process. Hence, the action of U reveals the three possible results associated with the output states which we labeled as \(\{|V\rangle |1\rangle ,\,|H\rangle |1\rangle ,\,|V\rangle |2\rangle \}\). Thus, detection at the corresponding output modes is the final step for implementing any POVM described by the operation given in Eq. (3). For FRIO discrimination, the unitary transformation in Eq. (3) allows us to cover all cases belonging to intervals I and II discussed in the last section. Indeed, detection in the \(|V\rangle |2\rangle\) mode corresponds to the inconclusive result for both states \(|\phi _i\rangle\) (see Fig. 1). On the other hand, detection in the \(|V\rangle |1\rangle\) mode corresponds to the success (error) in the discrimination of \(|\phi _{1(2)}\rangle\), while a detection in \(|H\rangle |1\rangle\) corresponds to the error (success) in the discrimination of \(|\phi _{1(2)}\rangle\).

For the case of interval III, the optimal measurement corresponds to a two-outcome projection61. In this case, the unitary transformation can be written as follows

$$\begin{aligned} U|\phi _1\rangle |1\rangle&=\sqrt{q_1}|V\rangle |1\rangle +\sqrt{r_1}|H\rangle |1\rangle ,\nonumber \\ U|\phi _2\rangle |1\rangle&=\sqrt{q_2}|V\rangle |1\rangle +\sqrt{p_2}|H\rangle |1\rangle , \end{aligned}$$
(6)

where there is no detection at \(|V\rangle |2\rangle\) mode. Thus, the detection in \(|V\rangle |1\rangle\) corresponds now with the inconclusive result. Then, for \(\eta _1<\eta _2\) in interval III we obtain the following optimal probabilities:

$$\begin{aligned} \begin{aligned} p_{2},r_{1}&= (\cos 2\theta \cos \theta _{3}\pm \sin 2\theta \sin \theta _{3})^{2},\\ q_{1},q_{2}&= (\cos 2\theta \sin \theta _{3}\pm \sin 2\theta \cos \theta _{3})^{2}.\\ \end{aligned} \end{aligned}$$
(7)

To detect the output photons, PerkinElmer single-photon avalanche detectors (APDs) were placed in the trigger path and the outputs of the Sagnac interferometer to record the photon statistics. A coincidence count module receives the signal from the detectors, where the timing delay was adjusted between each detector’s output and the heralding trigger signal. We actively control the pump laser power (1 mW), setting a 500 ps coincidence gate to minimize the accidental counts, generating a coincidence rate of \(\sim 1400\) photons pairs per second. This corresponds to a spectral brightness up to \(\sim 400000\) photon pairs (s mW nm)\(^{-1}\).

From the above discussion, it is clear that our experimental setup allows us to implement the optimal FRIO POVM for any pair of non-orthogonal states, arbitrary a priori probabilities, and any value of the fixed rate of inconclusive outcomes. This is done by adjusting the values of the angles \(\theta _1, \theta _2\), and \(\theta _3\) on the corresponding wave plates.

Results and discussion

Figure 2
figure 2

The success \(P_{s}\), error \(P_{e}\) and inconclusive Q average probabilities as a function of the inner product s, when \(\eta _{1}=\eta _{2}=0.5\) for both states \(\left| \phi _i\right\rangle\) are plotted in the third column. The first (second) column shows the success \(p_{1(2)}\), error \(r_{1(2)}\) and inconclusive \(q_{1(2)}\) rates when Alice prepares the state \(\left| \phi _{1}\right\rangle\) (\(\left| \phi _{2}\right\rangle\)) and Bob performs FRIO. The solid and dashed lines in each subplot represent the different values for the mentioned probabilities considering three different values for Q. They were computed using Eq. (8). As was aforementioned, for equally prepared states, we have only access to interval I, which is represented by the blue color in all subplots. Thus, since \(\eta _{1}=\eta _{2}\), each state’s success, error, and inconclusive rates behavior is the same while the inner product s varies.

Full size image
Figure 3
figure 3

The success \(P_{s}\) (\(p_{i}\)), error \(P_{e}\) (\(r_{i}\)) and inconclusive Q (\(q_{i}\)) probabilities as a function of the inner product s for preparation probabilities given by \(\eta _{1}=0.3\) and \(\eta _{2}=0.7\). Like Fig. 2, each line in the subplots represents the corresponding probabilities for the state in Eq. (8). Note that the curves in the first and second columns differ due to the preparation probabilities satisfying \(\eta _1\le \eta _2\). Therefore, the protocol tries to discriminate \(\left| \phi _{2}\right\rangle\) better than \(\left| \phi _{1}\right\rangle\). However, for optimizing the average success probability \(P_{s}\), the \(\left| \phi _{1}\right\rangle\) must also be considered. Additionally, the three optimization intervals appear for this case, and they are represented by the blue, green, and red colors.

Full size image

To show the practicality of the FRIO discrimination scheme using our experimental setup, we implemented the discrimination procedure between the two non-orthogonal polarization qubits given in Eq. (2) for a range of state overlaps s. A key feature of our experiment is that we use a single optical device to implement the complete range of FRIO from MED to UD and include any intermediate case. Indeed, the implementation of every strategy depends on the Sagnac interferometer configuration relying on the angles of the half-wave plates \(\hbox {HWP}_{{1}}\)(\(\theta _{1}\)), \(\hbox {HWP}_{{2}}\)(\(\theta _{2}\)) and \(\hbox {HWP}_{{3}}\)(\(\theta _{3}\)), as was shown in Eqs. (4), (5), and (7). To show the utility of the setup, we consider two cases regarding different states preparation probabilities, in which Alice can prepare the states among the cases \(\eta _{1}=\eta _{2}=0.5\) and \(\eta _{1}=0.3,\,\eta _{2}=0.7\) to experimentally validate Bob’s discrimination device. Since in the FRIO scheme these probabilities also bias the discriminated (detected) states [see Eq. (1)], Alice sets these probabilities experimentally by changing the integration time of the coincidence detection rate in terms of the settled values of \(\eta\) by the source. Moreover, as we mentioned before, we considered three different rates of inconclusive outcomes, namely \(Q\in \{0,Q_{0}/2,Q_{max}\}\), where the specific values depend on the inner product s and parameters \(\eta _i\). We recall that these three values of Q correspond to MED (\(Q=0\)), UD (\(Q=Q_{max}\)), and an intermediate case (\(Q=Q_0/2\)) in FRIO state discrimination.

In a preliminary step, with waveplates \(HWP_i(\theta _i)\) set to \(\theta _i=0\), we evaluated the polarization interference visibility \(\epsilon\) at the output of the interferometer by measuring in both Pauli bases \(\sigma _{z}\) and \(\sigma _{x}\), corresponding to the logical and diagonal polarization bases, respectively. We obtain the mean visibility \(\epsilon =0.981\pm 0.006\), which is typical in polarization-based experiments, where imperfections arise from experimental errors due to laser pump fluctuations in the SPDC process, imperfect spatial mode overlap, and misalignment in the waveplate settings. To take these into account in our comparison between experiment and theory, a white noise model is assumed, considering then the following state received by Bob

$$\begin{aligned} \rho _{i}= \epsilon \left| \phi _{i}\right\rangle \left\langle \phi _{i}\right| + \frac{(1-\epsilon )}{2}\mathbbm {1}, \end{aligned}$$
(8)

where \(i\in \{1,2\}\), \(\mathbbm {1}\) is the identity matrix.

FRIO quantum state discrimination was implemented for state overlap s ranging from orthogonal (\(s=0\)) to perfect overlap (\(s=1\)). As shown in Eq. (2), the inner product s is set experimentally through the angle of HWP(\(\theta\)), since \(s=\cos 4\theta\). For instance, for \(s=0\) we set \(\theta =\pi /8\) to generate \(|\phi _1\rangle\), and \(\theta =-\pi /8\) to generate \(|\phi _2\rangle\). For \(s=1\) we set \(\theta =0\) for both \(|\phi _1\rangle\) and \(|\phi _2\rangle\). The experimental results are shown in Fig. 2 (for the \(\eta _1=\eta _2=0.5\) case), and Fig. 3 (for the \(\eta _1=0.3\), \(\eta _2=0.7\) case). Alice used a data integration time of 10 s for both non-orthogonal states in Fig. 2, while for data in Fig. 3 times of 6 s and 14 s were used for \(\left| \phi _{1}\right\rangle\) and \(\left| \phi _{2}\right\rangle\), respectively. In both figures we plot the experimental average success \(P_{s}\), error \(P_{e}\), and inconclusive probability Q obtained from the recorded success \(p_i\), error \(r_i\), and inconclusive outcome probability \(q_i\) related to the state \(|\phi _i\rangle\), as shown in Eq. (1). These probabilities are presented as a function of seven different values for the inner product \(s\in {0,\ldots ,1}\). In Figs. 2 and 3, the first (second) column of plots corresponds to the results obtained when Alice sends the state\(\left| \phi _{1}\right\rangle\) (\(\left| \phi _{2}\right\rangle\)). The right-most column shows the average success, error, and inconclusive probabilities, defined as the sum of these probabilities considering \(\eta _{1}\) and \(\eta _{2}\) [see Eq. (1)]. Additionally, the expected (theoretical) values of the probabilities are defined in Eq. (1) using slightly mixed states (8) and are depicted by solid lines in every plot.

The error bars are smaller than the experimental points and were obtained with Gaussian error propagation and considering the Poisson statistic of the recorded coincidence counts. We can observe a good agreement between the expected and recorded results for every case regarding different inner product s. Additionally, for the case when \(\eta _1<\eta _2\) showed in Fig. 3, we plot the solid lines with three colors to indicate the three intervals in FRIO. Precisely, the blue color corresponds to interval I, while the green and red colors correspond to intervals II and III, respectively. Although the optimal measurement depends on the interval where the FRIO discrimination is performed, we obtain results close to the expected ones regarding the three intervals. This fact is a signature of the robustness of our platform against different experimental settings, which allow a wide range of discrimination processes contained in the FRIO scheme.

Conclusion

We present a single experimental device capable of discriminating between non-orthogonal polarization states of single photons in the fixed rate of inconclusive outcomes state discrimination scheme, for which the well-known Minimum Error and Unambiguous state discrimination methods are limiting cases. The device is based on a polarization-controlled Sagnac interferometer with nested waveplates and allows for FRIO state discrimination to be implemented for a wide range of parameters, which are defined by the overlap and input statistics of the input states tested. In our setup, Alice controls a photon pair source and encodes non-orthogonal polarization states into a heralded single photon. She sends the photon to Bob, who uses the interferometer to implement a POVM measurement with up to three outcomes. We show that this single device can implement a wide range of state discrimination procedures. Good agreement between theory and experimental results is obtained considering the two-path Sagnac interferometer visibility, which is about \(98\%\). The versatility of the single device makes it directly applicable for quantum information tasks such as quantum communications.