## Abstract

Quantum mechanics and gravitation are two pillars of modern physics. Despite their success in describing the physical world around us, they seem to be incompatible theories. There are suggestions that one of these theories must be generalized to achieve unification. For example, Born’s rule—one of the axioms of quantum mechanics—could be violated. Born’s rule predicts that quantum interference, as shown by a double-slit diffraction experiment, occurs from pairs of paths. A generalized version of quantum mechanics might allow multipath (i.e., higher-order) interference, thus leading to a deviation from the theory. We performed a three-slit experiment with photons and bounded the magnitude of three-path interference to less than 10^{−2} of the expected two-path interference, thus ruling out third- and higher-order interference and providing a bound on the accuracy of Born’s rule. Our experiment is consistent with the postulate both in semiclassical and quantum regimes.

Born’s interpretation (*1*) of the wave function ψ(**r**, *t*) for a quantum mechanical state stipulates that the probability density to find a particle at position **r** and at time *t* is given by**r** after passing through an aperture with two slits, *A* and *B*, is given by*P _{i}* to be the probability with only slit

*i*(

*i*=

*A*,

*B*) open. The corresponding (second-order) interference term can be defined as

*A*,

*B*, and

*C*(Fig. 1), we find

*2*). These possibilities can be associated with any degree of freedom, such as spatial paths, energetic states, angular momentum states, etc. Even if multiple particles are involved, interference occurs in pairs of possibilities. Consequently, we define the third-order interference term

*I*for a three-path configuration (mutually exclusive) as the deviation of

_{ABC}*P*from the sum of the individual probabilities and the second-order interference terms:

_{ABC}*2*).

The nonzero interference term *I _{AB}* is expected in all wave theories, including quantum mechanics (

*3*,

*4*). The next higher-order (i.e., three-path) interference term

*I*will be zero in all wave theories, with a square-law relation between the field energy (or probability density) and field amplitude, which is the case in quantum mechanics with Born’s rule. Moreover, if there is no interference at a certain level in the hierarchy, the higher-order terms must vanish as well (

_{ABC}*2*).

Our aim is to establish experimentally whether the value of *I _{ABC}* is different from zero. We measure all seven probability terms of Eq. 5 plus the probability

*P*

_{0}of detecting particles when all slits are closed.

*P*

_{0}represents the probability of the empty set in an abstract definition, or a background signal in the experiment. The eight terms are obtained by sending optical photons through three slits, which can be opened or closed individually (see Fig. 1 for the slit details and Fig. 2 for the setup). A double-slit experiment could be used to test Born’s rule, but then one would have to measure the nonzero double-slit interference term and compare it with the theoretical prediction. This would be sensitive to experimental parameters such as slit dimensions, wavelength of incident photons, and distance between detector and slits, each with its attendant error. In contrast, we expect the three-path interference term

*I*to be zero, with the advantage of being independent of many experimental parameters, thus enabling a more precise null test for Born’s rule.

_{ABC}We measure the terms in Eq. 5 as well as *P*_{0}, which accounts for the inevitable detector noise and background signal. The measured quantity ε based on Eq. 5 is given by

To measure κ in various optical power regimes, we used different types of photon sources. Figure 2 shows details of the experimental setup. We used a laser attenuated to a power level of a few microwatts down to ~200 fW (single-photon level) as well as heralded single photons (~40,000 photons/s) created by spontaneous parametric downconversion (*5*).

At the photon-counting level, the detection mechanism is based on a silicon avalanche photodiode (APD), and thus the particle-like nature of light is incorporated in the experiments. At the microwatt level, a series of measurements was performed with a standard optical power meter, using a silicon photodiode. The power meter measurements investigated the optical regime in which particle character is not of concern. In all cases we performed a large number of measurements at fixed points in the diffraction pattern [fig. S1 in (*5*)]. In addition, we have also performed measurements to check the variation of κ as a function of detector position. Born’s rule would predict that κ should be independent of detector position. However, systematic errors may vary with the position and therefore are seen to bring a variation in the measured value of κ at different detector positions even in our experiment. Nonetheless, the mean κ is within the bounds set by the attendant errors at each such detector position.

The typical distributions of measured values of κ are shown in Fig. 3, with photon streams from a laser attenuated to different levels (Fig. 3, A and B) and from a heralded single-photon source (Fig. 3C). κ is calculated from the measured interference intensities for the eight independent slit combinations at a fixed position. The order of the eight slit combinations was chosen randomly for reducing systematic influences on κ caused by slow variations of the photon flux. Each combination in a run was measured for a certain photon-count integration time, and up to 100 runs were cycled to obtain a statistically significant sample of κ values. Among the many positions in the diffraction pattern, we chose the central maximum of the triple-slit combination (yielding the maximum number of coincidence photon counts) to obtain our data (*5*). For the single-photon source, we measured at each slit combination until the trigger count reached 30 million, which was a good compromise between accumulating a statistically significant number of coincidences for the different slit combinations and ensuring a low drift of the photon source between measurements.

With a null experiment, a very careful analysis of random and systematic errors must be undertaken, as our bound on the amount of three-path interference will be directly related to the level of experimental uncertainty. Among the random errors in our setup, thermal and acoustic fluctuations cause the source fluxes to vary in time. In addition, detection efficiency and optical alignment can change. In particular, there will be some mechanical vibration of the thin (25 μm) slotted steel membrane apertures, causing a variable slit transmission due to near-field diffraction. In addition, for power meter measurements, the instrumental error is added to the above error sources, whereas for photon counting, the Poissonian distributed counting error is the dominating fluctuation. Because of the random nature of the individual errors, we used Gaussian error propagation to estimate the error of κ, where we used the standard variances of the individual measurement values calculated from a large number of repetitions of the experiments. In some cases where we observed a drift in the rates, we found the Allen variance of the values to be a better estimator for error propagation. This is justifiable because κ is calculated from eight measurements taken in direct succession, and the variance between subsequent samples of each quantity *p _{A}*,

*p*, etc., is therefore the most suitable error estimator.

_{B}Once we understand the random errors, we can characterize the systematic errors. Our experiment and the measurement of κ are convenient, as they neither require the slits to be identical nor require the transmission values to be perfectly 1 and 0. On the other hand, what matters is the absence of correlation or systematic variation in how the slits behave while switching between slit combinations. Note that the size of the slits and the wavelength make independent shutters difficult to insert, and we used a static opening mask plate in front of the actual slits for blocking and unblocking the individual slits.

Our approach can potentially introduce unwanted correlations between the switching of different combinations. This occurred in our case; a fault in the blocking mask in the *BC* combination caused opening *B* to be shifted off its nominal position by 8 μm. At zero distance between opening and slit masks, this shift would not affect the transmittance of the diffracting slit. However, with the finite separation of 50 μm between the two masks, it does matter. Using a two-dimensional finite-difference time domain simulation of the light field between the wider opening and narrower slit, we found that this lateral misalignment could change the effective slit transmission by ±3%, depending on the relative position. Using this to adjust the slit transmittance value for slit *B* in the combination *BC* leads to a value of |κ| = 0.01. Bringing the masks closer would reduce this error, but the thin membranes tend to stick together when they come closer than 50 μm.

Another major systematic error is detector nonlinearity. There is no perfect detector, and efficiency and nonlinearity will always be finite. For our commercial power meter, the nonlinearity is 0.5% for all ranges. For the worst case, at the central maximum of the diffraction pattern, this results in a systematic error of |κ| = 0.003. For photon-counting detectors, there is always a finite dead-time during which they are blind to photons. Depending on the flux, this results in a saturation effect; that is, the detector response does not follow the square law, but has a deviation potentially indistinguishable from a violation of Born’s rule. Therefore, to keep the nonlinearity error negligible, we used count rates below 100,000/s.

Combining the various error sources, our particular setup enables us to bound the measurements of κ to better than 0.01, thus providing a bound on the accuracy of Born’s rule. This is in good agreement with the values measured for single photons, κ = 0.0064 ± 0.0120, and for the attenuated laser beams, κ = 0.0073 ± 0.0018 (power meter measurement) and κ = 0.0034 ± 0.0038 (APD measurement) from Fig. 3.

Note that any significant nonzero observation of *I _{ABC}* would imply that Born’s rule does not strictly hold. The consequences of detecting even a small amount of three-way interference (by deviating from the quantum mechanical null prediction) would be tremendous. A modification to Born’s rule that leads to multi-order interference would have repercussions on the allowable dynamics. In particular, if probability must be conserved, then Schrödinger’s equation would likely have to be modified as well. Nonlinear extensions to quantum mechanics are one way to generalize it (

*6*–

*8*), and there have been efforts to test these nonlinearities (

*9*–

*12*). However, in such experiments, a model was assumed for the nonlinear variant of the Schrödinger equation, and efforts were concentrated on estimating the coefficient of the nonlinearity. In contrast, we present a dedicated test for Born’s rule and are able to confirm it within our experimental limitations without depending on specific nonlinear extensions of quantum mechanics.

We are able to bound the magnitude of the third-order interference term to less than 10^{−2} of the regular expected second-order interference, at several detector positions. Thus, our experiment is able to rule out the existence of third-order interference terms (and, in effect, any higher-order interference terms) up to this bound. This bound on the accuracy of Born’s rule is relevant for theoretical attempts to derive it (*13*) as well as for the generalization of quantum mechanics. Among the consequences of such a generalized theory, it would require more detailed specifications of the quantum system (*14*) and could modify computational complexity by allowing one to distinguish between orthogonal states, thus breaking quantum cryptography and making quantum computing more powerful [i.e., super-quantum computing (*15*)]. The triple-slit experiment is a simple and very natural system to investigate three-path interference, but it is not the only possible implementation; any configuration of three mutually exclusive quantum paths can be used in such a test. Although our implementation did not lead to the observation of any deviations, such future tests with other systems might indeed lead to improved accuracies on the bound of κ. It would be interesting to perform tests with other types of particles such as neutrons (*16*, *17*), tests of C_{60} molecule interference (*18*) or electron interference with slits or potential wells (*19*, *20*), or tests of a system with a global wave function, such as a Bose-Einstein condensate (*21*, *22*).

## Supporting Online Material

www.sciencemag.org/cgi/content/full/329/5990/418/DC1

Materials and Methods

Fig. S1

References

## References and Notes

- ↵
- ↵
- ↵
- ↵
- ↵
See supporting material on
*Science*Online. - ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- U.S., C.C., and T.J. performed the experiments; C.C., R.L., and G.W. conceived of and supervised the experiment and developed the theoretical aspects; G.W. and U.S. performed the modeling and the data and error analysis; and all authors contributed to the writing of the paper. We thank H. Hübel, R. Kaltenbäck, A. J. Leggett, A. Sinha, R. D. Sorkin, and A. Zeilinger for valuable discussions, and Z. Medendorp and I. Söllner for technical assistance earlier in the experiments. Supported by The Natural Science and Engineering Research Council, Canada Foundation for Innovation, European Research Area, Quantum Works, Ontario Centres of Excellence, and Canadian Institute for Advanced Research.