SUPPLEMENTARY INFORMATION

SUPPLEMENTARY INFORMATION doi: 0.08/nPHYS777 Optical one-way quantum computing with a simulated valence-bond solid SUPPLEMENTARY INFORMATION Rainer Kaltenbaek,, Jonathan Lavoie,, Bei Zeng, Stephen D. Bartlett, and Kevin J. Resch Institute for Quantum Computing and Department of Physics & Astronomy, University of Waterloo, Waterloo NL G, Canada Institute for Quantum Computing and the Department of Combinatorics and Optimization, University of Waterloo, Waterloo NL G, Canada School of Physics, The University of Sydney, Sydney, New South Wales 006, Australia I. EXPERIMENTAL SETUP The entangled photon pairs in our experiment are generated using two separate type-i spontaneous parametric down conversion (SPDC) sources [, ]. Each consists of a pair of mm thick beta-barium-borate (β-bab O, or BBO) crystals, their optical axes oriented perpendicular to each other. Longitudinal walk off in the SPDC crystals is compensated using a 0.5mm quartz, a mm quartz and a mm α-bbo crystal before the first SPDC source, and a mm α-bbo and a mm quartz crystal between the two sources. Additional transverse walk-off is compensated by placing mm thick BiB O 6 crystals cut at = 5.6 and φ =0 in modes and (see Figure a). These angles are chosen such that the crystals compensate transverse walk-off without introducing longitudinal walk off. All photons pass through nm FWHM bandwidth filters. The phases in the setup and the polarisation rotation in the single-mode fibres is set such that the sources produce singlet states in modes & and &. In modes & we measure a fidelity of (96.9 ± 0.5)% with the ideal singlet state ψ = ( HV VH ), and a tangle of 0.9 ± 0.0. For the second source the fidelity is (96.9 ± 0.6)% and the tangle is 0.90 ± 0.0. We had single count rates of about 00kHz in the qubit analysers and (see Figure a), and single count rates of around 80kHz in the detectors D and D in the qutrit analyser (see Figure c) (both sources contribute to these latter single count rates). The two-fold coincidence count rate for the first source was 7.kHz between qubit analyser and D in the FIG. : Experimental Setup. (a) Two type-i SPDC sources are used to generate entangled pairs. Longitudinal and transverse walk-off are compensated via combinations of birefringent crystals T, T and C. All photons are coupled into single-mode fibers. Polarisation rotation in the fibres is compensated via polarisation controllers, and the phase is adjusted by tilting quarter waveplates (QWP). Photons in modes and are fed into a 50:50 beam splitter (BS). Detecting two photons in one of the BS outputs projects these photons on the symmetric qutrit subspace. They can then be treated as a qutrit. (b) Each qubit analyser consists of a half waveplate (HWP), a QWP and a polarising BS. Both outputs of the polarising BS are monitored. (c) In the qutrit analyser the biphoton forming the qutrit is probabilistically split up using a BS. Phases introduced at the BS are compensated by tilted QWPs. A combination of QWPs and HWPs allows to project the qutrit onto a state of our choice. A successful projection is heralded by a coincidence event between detectors D and D. These authors contributed equally to this work. nature physics www.nature.com/naturephysics

Rot. R x() R y() R z() plus minus id α χ α χ α χ α χ α χ α χ + 0 0 0 TABLE : Analyser parameters for qutrit measurements for rotation gates. In the qutrit analyser, one photon is projected on ψ, the second one on ψ, where ψ m = cos α m H + e iχm sin α m V. Each qutrit measurement has three possible outcomes, plus, minus and id corresponding to three different sets of parameters for the qutrit analyser. We provide the settings for rotations around each of the Cartesian axes. 0 + 0 0 0 + 0 + +i 0 i 0 0 +i α 0 0 0 0 0 0 χ 0 0 0 0 0 0 0 0 i +i i 0 0 0 0 α 0 ξ ξ η η 0 ξ ξ η η χ 0 0 0 0 0 p TABLE : Analyser parameters for qutrit measurements used for over-complete tomography. Each qutrit measurement is implemented by splitting the biphoton probabilistically at a beam splitter and projecting the two photons q (m =, ) on ψ m = cos α m H + e iχm sin α m V. For brevity, we use the definitions ξ = arccos( ) and η = arccos( ), and p denotes the success probability for a given qutrit projection. These success probabilities are taken into account by our tomographic reconstruction of the density matrix. qutrit analyser. For the second source we had a two-fold coincidence count rate of 5.9kHz between qubit analyser and D in the qutrit analyser. One photon of each pair is measured directly at the source, using polarisation analysers. The modes for both measurement outcomes are coupled into single-mode fibres and monitored via single-photon detectors (Perkin-Elmer, SPCM-QC). The two remaining photons are coupled into single-mode fibres and sent to a quantum interferometer and analyser setup. The input modes of the interferometer are overlapped at a 50:50 beam splitter (BS). If the input photons are set to have the same polarisation, Hong-Ou-Mandel (HOM) interference [] occurs. Postselecting on four-fold events with one photon in mode, one in mode and two photons in the output mode of the BS indicated in Figure a, we observe constructive HOM interference with a visibility of 95.7 ±.7%. If the polarisation of the photons in modes and is not projected on a polarisation product state, it will depend on the biphoton state whether the two photons leave through the same or different BS outputs []. In particular, coincidence detection events between two different BS outputs only occur for the two-photon singlet state. By post-selecting on a biphoton excitation in one output mode of the BS, the biphoton is projected onto a symmetric subspace and can be described as a qutrit [5, 6]. This is identical to the symmetrisation needed to generate an AKLT state [7] (see Fig. in the main text). We measure this qutrit using the analyser outlined in Figure c. The analyser works by probabilistically splitting up the biphoton at a BS, and by performing a qubit projective measurement on each of the output modes and of the BS. We project on a given qutrit state by projecting the two photons on corresponding qubit states. In particular, the photon in mode m =, of the qutrit analyser is projected on state ψ m = cos α m H + e iχm sin α m V. A successful projective measurement of a qutrit is heralded by a coincidence event between detectors D and D. To calculate which qutrit state such a coincidence event signals we can propagate our two-qubit state ψ ψ back through the beam splitter. All contributions in the unused of the two input ports of the BS can be neglected. The (unnormalised) two-photon state then becomes: cos α cos α a H a H + eiχ cos α sin α +e iχ cos α sin α a H a V +ei(χ+χ) sin α sin α a V a V vac. () In our qutrit basis we can write this as: cos α cos α + eiχ cos α sin α +e iχ cos α sin α 0 + e i(χ+χ) sin α sin α. () In general, the success probability of this projective qutrit measurement depends on the qutrit state. For example, to project on the biphoton state a H a H vac, we set both analysers to H. Given that the biphoton is in the correct nature physics www.nature.com/naturephysics

0 + 0 0 0 + 0 + +i 0 i 0 0 +i 0 i +i i H 8 595 7 7 9 9 58 8 7 0 8 98 5 5 V 7 95 7 07 0 5 6 7 0 8 H + 8 80 7 6 68 6 96 90 86 75 57 7 76 00 9 9 0 8 7 8 88 67 0 78 60 8 86 R 90 5 78 6 7 58 95 69 80 6 7 06 09 7 L 0 97 7 6 65 70 6 0 8 7 66 88 0 H 7 7 7 86 78 77 6 65 98 7 80 V 59 0 0 5 5 5 77 5 8 V + 56 79 8 9 77 5 8 77 6 50 5 60 76 7 68 0 5 5 0 66 66 5 5 9 6 R 9 6 5 0 9 57 6 87 8 6 8 90 89 L 99 7 8 5 5 8 6 9 9 6 8 59 H 5 68 8 6 0 0 8 58 85 58 60 58 85 V 8 08 9 55 96 80 56 60 70 67 70 6 + + 7 7 9 8 5 9 7 8 6 55 69 0 68 85 8 6 0 79 5 5 56 7 8 R 6 85 9 7 5 56 7 8 5 8 9 5 L 88 05 7 80 6 75 8 60 9 66 5 9 0 H 8 6 0 80 5 8 67 66 5 68 9 5 66 79 V 5 87 5 0 6 67 6 99 98 7 77 57 80 0 + 5 8 7 67 50 5 56 6 67 9 68 58 66 95 66 7 69 7 8 86 5 67 8 59 9 0 75 89 R 89 87 8 66 7 6 80 9 56 59 0 6 57 L 5 89 67 5 6 79 79 9 96 9 0 5 0 5 H 0 58 70 67 9 77 8 9 5 00 96 80 V 98 85 5 6 9 59 7 98 68 9 9 76 65 7 60 R + 6 68 69 0 6 8 67 99 5 5 0 9 7 59 9 0 6 57 8 9 7 9 55 68 8 8 R 5 6 8 7 8 56 9 7 8 55 5 96 87 89 L 6 9 57 6 9 68 6 8 5 66 96 6 57 H 9 6 9 60 7 97 66 56 5 67 7 7 V 0 7 6 65 8 58 67 8 86 9 0 86 5 99 6 L + 85 67 6 9 6 00 7 8 6 9 7 95 0 8 67 9 9 0 97 80 7 7 0 8 R 80 6 75 57 7 66 8 00 5 6 68 9 L 5 7 9 50 6 0 7 06 5 8 8 TABLE : Table of tomography results. We performed a tomographically-overcomplete set of measurements on our qubitqutrit-qubit state. The states in the first column indicate the measurement settings for the first qubit, the states in the second column indicate the settings for the other qubit, and the states in the top row denote the qutrit measurement settings. All counts given are raw four-fold coincidence events integrated over 80s. For the tomographic reconstruction, one has to take into account the fact that the qutrit measurement only works with a certain probability (see Table ). state, a coincidence will occur with probability. As a second example, in order to project on the biphoton state a H a V vac, we choose ψ = H and ψ = V. If the biphoton is in the right state, the success probability will only be because the photons can be split in four possible ways, and only one leads to a coincidence event. Table lists the parameters α m and χ m we have to choose to project on a qutrit measurement for a rotation by an angle around axis ˆx, ŷ or ẑ. The probability for each of these qutrit measurements to work is. By taking into account the V outcomes of the polarising beam splitters (PBSs) in the qutrit analyser, this probability could be increased to. In order to analyse the qubit-qutrit-qubit state generated in our setup we performed a tomographically over-complete set of measurements and performed tomography using a maximum likelihood technique [8, 9]. The set of measurements performed for the two qubits was H, V, +,, R, and L. In Table we list the set of qutrit measurements as well as the corresponding parameters α m and χ m for the settings of the qubit measurements in the qutrit analyser. The table also shows the probabilities with which each of the projections succeeds given that the biphoton is in the corresponding qutrit state. A qubit analyser where both outputs ( H and V ) are monitored can be operated in various ways. One is to nature physics www.nature.com/naturephysics

outcome plus logical input state H V h + h m + m 0 0.85 ± 0.0 0.9 ± 0.0 0.97 ± 0.0 0.99 +0.0 0.0 0.9 ± 0.0 0.95 ± 0.0 /8 0.98 +0.0 0.0 0.87 ± 0.05 0.97 ± 0.0 0.9 ± 0.06 0.95 ± 0.0 0.95 ± 0.0 / 0.85 ± 0.05 0.9 ± 0.0 0.9 ± 0.05 0.88 ± 0.05 0.98 ± 0.0 0.9 ± 0.0 /8 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.95 ± 0.0 0.86 ± 0.0 / 0.9 ± 0.0 0.8 ± 0.06 0.9 ± 0.0 0.9 ± 0.05 0.9 ± 0.0 0.78 ± 0.0 / 0.88 ± 0.05 0.9 ± 0.06 0.9 ± 0.05 0.90 ± 0.06 0.86 ± 0.0 0.9 ± 0.0 0.97 ± 0.0 0.96 ± 0.0 0.9 ± 0.0 0.89 ± 0.0 0.88 ± 0.0 0.87 ± 0.0 5/ 0.9 ± 0.0 0.9 ± 0.0 0.8 ± 0.0 0.9 ± 0.0 0.95 ± 0.0 0.96 ± 0.0 / 0.9 ± 0.0 0.95 ± 0.0 0.9 ± 0.0 0.99 +0.0 0.0 0.86 ± 0.0 0.96 ± 0.0 7/ 0.9 ± 0.0 0.9 ± 0.0 0.89 ± 0.05 0.97 ± 0.0 0.9 ± 0.0 0.90 ± 0.0 0 0.97 ± 0.0 0.9 ± 0.0 0.87 ± 0.05 0.8 ± 0.06 0.95 ± 0.0 0.89 ± 0.0 /8 0.95 ± 0.0 0.98 +0.0 0.05 0.8 ± 0.0 0.8 ± 0.05 0.9 ± 0.0 0.9 ± 0.0 / 0.9 ± 0.0 0.99 +0.00 0.09 0.96 ± 0.0 0.9 ± 0.05 0.95 ± 0.0 0.98 ± 0.0 minus /8 0.9 ± 0.0 0.9 ± 0.0 0.99 +0.0 0.0 0.9 ± 0.0 0.95 ± 0.0 0.89 ± 0.0 / 0.90 ± 0.0 0.96 ± 0.0 0.89 ± 0.05 0.9 ± 0.05 0.9 ± 0.0 0.89 ± 0.0 / 0.998 +0.00 0.06 0.97 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.97 ± 0.0 0.98 ± 0.0 0.90 ± 0.0 0.9 ± 0.0 0.86 ± 0.0 0.96 ± 0.0 0.9 ± 0.0 0.88 ± 0.0 5/ 0.9 ± 0.0 0.98 ± 0.0 0.86 ± 0.0 0.89 ± 0.0 0.95 ± 0.0 0.9 ± 0.0 / 0.9 ± 0.0 0.95 ± 0.0 0.9 ± 0.0 0.87 ± 0.0 0.89 ± 0.0 0.90 ± 0.0 7/ 0.95 ± 0.0 0.98 +0.0 0.05 0.9 ± 0.0 0.8 ± 0.05 0.95 ± 0.0 0.9 ± 0.0 id n.a. 0.90 ± 0.0 0.9 ± 0.0 0.89 ± 0.0 0.96 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 TABLE : Fidelities for rotations around the ˆx axis. This table lists the fidelities for various input states of the computation compared with the theoretical expectation given an ideal AKLT state. These fidelities are given for a number of angles of rotation around the ˆx axis. We list them separately for the three distinct outcomes plus, minus and id of the qutrit measurement. For each input state we measured the id outcome only once because it is independent of the rotation angle. adjust the coupling efficiencies of the two single-mode fibre couplers such that they are approximately the same. This method is, however, very susceptible to long-time drifts of the sources because these might again lead to an unbalance between the two couplers. The method we chose is the following. Let us assume we want to measure the counts for a projection on H. Then we record the counts in both PBS outputs once with the waveplates set such that the transmitted path in the PBS corresponds to H, and we record the counts in both PBS outputs for the same amount of time with the waveplates set such that the transmitted path corresponds to V. We do the same for any polarisation we project on, and by adding up the respective counts we average over any imperfections in the balance of the coupling efficiencies. If we have N qubit analysers for which we apply this technique, the number of combinations to measure will be N. For our tomography measurements of the AKLT state we used this technique for modes and. For the single-qubit rotation measurements we used it for mode, but we only monitored the transmitted mode in analyser in order to simplify the software implementation of the scan routine. II. RESULTS All measured counts correspond to four-fold coincidence detection events between one detector in each of the two qubit analysers and the two detectors D and D in the qutrit analyser (see Fig. ). Table lists the fourfold coincidence counts measured for all tomographical settings. For each setting of the analyser waveplates we integrated over 60s. Because we monitored both outputs in each of the two qubit analysers and applied the technique described above in order to average over any unbalance between the two analyser outputs, we have to measure combinations of settings per projective measurement. This results in 0s overall measurement time per projective measurement. In order to reduce the effect of slow drifts in the setup, we performed the full set of measurements twice, in each case randomly ordering the settings, resulting in a measurement time of 80s per setting in Table. nature physics www.nature.com/naturephysics

outcome plus minus logical input state H V h + h m + m 0 0.95 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.88 ± 0.0 0.9 ± 0.0 0.96 ± 0.0 /8 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.86 ± 0.0 / 0.9 ± 0.0 0.90 ± 0.0 0.9 ± 0.0 0.88 ± 0.0 0.9 ± 0.0 0.98 ± 0.0 /8 0.9 ± 0.0 0.95 ± 0.0 0.95 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.87 ± 0.0 / 0.90 ± 0.0 0.9 ± 0.0 0.85 ± 0.0 0.88 ± 0.0 0.8 ± 0.0 0.88 ± 0.0 / 0.9 ± 0.0 0.86 ± 0.0 0.85 ± 0.0 0.87 ± 0.0 0.90 ± 0.0 0.89 ± 0.0 0.8 ± 0.0 0.85 ± 0.0 0.87 ± 0.05 0.95 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 5/ 0.8 ± 0.0 0.88 ± 0.0 0.90 ± 0.0 0.95 ± 0.0 0.98 ± 0.0 0.95 ± 0.0 / 0.87 ± 0.0 0.89 ± 0.0 0.88 ± 0.0 0.9 ± 0.0 0.96 ± 0.0 0.9 ± 0.0 7/ 0.9 ± 0.0 0.90 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.998 +0.00 0.09 0 0.90 ± 0.0 0.87 ± 0.0 0.90 ± 0.0 0.86 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 /8 0.9 ± 0.0 0.89 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.99 +0.0 0.0 0.98 ± 0.0 / 0.9 ± 0.0 0.90 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 /8 0.9 ± 0.0 0.90 ± 0.0 0.9 ± 0.0 0.90 ± 0.0 0.9 ± 0.0 0.96 ± 0.0 / 0.9 ± 0.0 0.9 ± 0.0 0.96 ± 0.0 0.85 ± 0.0 0.9 ± 0.0 0.96 ± 0.0 / 0.9 ± 0.0 0.96 ± 0.0 0.9 ± 0.0 0.89 ± 0.0 0.96 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.90 ± 0.0 0.9 ± 0.0 0.96 ± 0.0 0.95 ± 0.0 5/ 0.95 ± 0.0 0.996 +0.00 0.00 0.90 ± 0.0 0.85 ± 0.0 0.9 ± 0.0 0.96 ± 0.0 / 0.9 ± 0.0 0.87 ± 0.0 0.85 ± 0.0 0.8 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 7/ 0.8 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.87 ± 0.0 0.95 ± 0.0 id n.a. 0.9 ± 0.0 0.88 ± 0.0 0.89 ± 0.0 0.9 ± 0.0 0.96 ± 0.0 0.9 ± 0.0 TABLE 5: Fidelities for rotations around the ŷ axis. In the main text we gave the measurement results for the rotation of a logical input state H around all three coordinate axes. For completeness, we have done the same measurements for a set of logical input states. In particular, we prepared input states H, V, ±, h ±, and m ±. Here, ± =( H ± V )/, h + = cos( 8 ) H + sin( 8 ) V and h = sin( 8 ) H + cos( 8 ) V are the eigenstates of the Hadamard operator, m+ = cos( ξ ) H +ei sin( ξ ) V is the magic state [0] with ξ = arccos( ), and m m + = 0. For the input states ± we only performed a rotation around the Cartesian ẑ axis. For the other six input states we performed rotations around all three Cartesian axis. It is important to note, that in our setup it is possible to rotate an arbitrary input state around any of the three coordinate axes because we project the first qubit on a given state rather than on a measurement basis. As we mentioned in the main text we prepare a logical input state ψ by projecting on the qubit state orthogonal to it, i.e. ψ. Because we post select on four-fold coincidence events, we automatically disregard those cases where the outcome of our measurement would be ψ. If one took into account both outcomes of the projective measurement, each of them would occur randomly with probability, and if we obtained the outcome ψ, the logical input state would be the state orthogonal to what we wanted to prepare. To correct for this error for any arbitrary state is impossible because that would require a universal-not operation, which is non-unitary []. For deterministic operation, one is restricted to input states such that errors in the preparation can be corrected via Pauli operations, i.e. any of the states along the coordinate axes of the Bloch sphere. For each input state, rotation angle, qutrit outcome and rotation angle we performed a tomographically overcomplete set of measurements ( H, V, +,, R, and L ) on the qubit carrying the result of our single-qubit logic gate. The measurement times over which we collected data varied because the count rates dropped continuously due to long-time drifts in the setup. For the input states H, V, and h ±, the measurement time was 0s per setting. Because we applied our technique of summing recurring occurrences of identical polarizer settings for the two outputs of the polarising beam splitter in qubit analyser (see Fig. ) the effective measurement time was twice as long, i.e. 0s. For the other input states we doubled this scan time to 80s per analyser setting. In Tables, 5, and 6 we list the fidelities of the reconstructed maximum-likelihood density matrices for all rotations of various input states compared with what we would expect given a perfect AKLT state. Table shows these fidelities for rotations R x () around the ˆx axis by an angle. For each of the 6 different logical input states nature physics www.nature.com/naturephysics 5

outcome plus minus logical input state H V h + h m + m + 0 0.88 ± 0.0 0.95 ± 0.0 0.90 ± 0.0 0.90 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.90 ± 0.0 /8 0.9 ± 0.0 0.9 ± 0.0 0.87 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.98 +0.0 0.0 0.99 ± 0.0 / 0.85 ± 0.0 0.86 ± 0.0 0.9 ± 0.05 0.90 ± 0.0 0.95 ± 0.0 0.95 ± 0.0 0.98 +0.0 0.05 0.90 ± 0.05 /8 0.86 ± 0.0 0.96 ± 0.0 0.89 ± 0.05 0.90 ± 0.0 0.87 ± 0.05 0.9 ± 0.0 0.9 ± 0.0 0.89 ± 0.05 / 0.9 ± 0.0 0.9 ± 0.0 0.79 ± 0.06 0.98 ± 0.0 0.85 ± 0.05 0.87 ± 0.0 0.88 ± 0.05 0.90 ± 0.0 / 0.87 ± 0.0 0.9 ± 0.0 0.95 ± 0.0 0.9 ± 0.0 0.8 ± 0.05 0.89 ± 0.0 0.9 ± 0.05 0.88 ± 0.06 0.9 ± 0.0 0.87 ± 0.0 0.80 ± 0.06 0.8 ± 0.05 0.9 ± 0.0 0.9 ± 0.0 0.95 ± 0.0 0.96 ± 0.0 5/ 0.9 ± 0.0 0.88 ± 0.0 0.99 +0.0 0.0 0.9 ± 0.05 0.89 ± 0.0 0.996 +0.00 0.05 0.9 ± 0.0 0.96 ± 0.0 / 0.9 ± 0.05 0.96 ± 0.0 0.9 ± 0.05 0.90 ± 0.06 0.95 ± 0.0 0.89 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 7/ 0.95 ± 0.0 0.9996 +0.000 0.078 0.9 ± 0.07 0.998+0.00 0.00 0.9 ± 0.0 0.97 ± 0.0 0.8 ± 0.06 0.96 ± 0.0 0 0.88 ± 0.0 0.96 ± 0.0 0.89 ± 0.0 0.89 ± 0.0 0.90 ± 0.0 0.99 +0.0 0.0 0.90 ± 0.0 0.9 ± 0.0 /8 0.89 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.97 +0.0 0.05 0.95 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.88 ± 0.05 / 0.96 ± 0.0 0.90 ± 0.0 0.9 ± 0.0 0.90 ± 0.05 0.97 ± 0.0 0.97 ± 0.0 0.89 ± 0.05 0.95 ± 0.0 /8 0.9 ± 0.0 0.9 ± 0.0 0.97 ± 0.0 0.9 ± 0.05 0.90 ± 0.0 0.9 ± 0.0 0.9 ± 0.05 0.9 ± 0.0 / 0.90 ± 0.0 0.89 ± 0.0 0.96 ± 0.0 0.85 ± 0.06 0.90 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 / 0.9 ± 0.0 0.9 ± 0.0 0.99 +0.00 0.08 0.9 ± 0.0 0.9 ± 0.0 0.99 ± 0.0 0.98+0.0 0.05 0.9999 +0.0000 0.0 0.86 ± 0.0 0.87 ± 0.0 0.9 ± 0.0 0.9 ± 0.05 0.90 ± 0.0 0.9 ± 0.0 0.9 ± 0.0 0.90 ± 0.05 5/ 0.88 ± 0.05 0.99 +0.00 0.0 0.95 ± 0.05 0.90 ± 0.06 0.90 ± 0.0 0.9 ± 0.0 0.9 ± 0.05 0.8 ± 0.06 / 0.97 ± 0.0 0.9 ± 0.0 0.8 ± 0.09 0.98 ± 0.0 0.80 ± 0.05 0.86 ± 0.0 0.97 ± 0.0 0.97 ± 0.0 7/ 0.99 +0.0 0.0 0.996 +0.00 0.08 0.9 ± 0.05 0.87 ± 0.06 0.9 ± 0.0 0.8 ± 0.0 0.89 ± 0.05 0.9 ± 0.05 id n.a. 0.97 ± 0.0 0.97 ± 0.0 0.96 ± 0.0 0.97 ± 0.0 0.9 ± 0.0 0.98 ± 0.0 0.90 ± 0.0 0.90 ± 0.0 TABLE 6: Fidelities for rotations around the ẑ axis. and each angle we observed the plus as well as the minus outcome of the qutrit measurement. Because the id outcome is independent of the rotation angle, we only measured it once for every rotation axis. In order to estimate the error of these fidelities, we performed Monte-Carlo simulations on the measured counts, which are assumed to be the mean of a Poissonian count distribution, to reconstruct a set of 00 density matrices for each result. Where a single error is given, it corresponds to the standard deviation of the fidelities of these 00 density matrices. Because the fidelity cannot exceed one, those fidelities very close to have asymmetric error bars. To calculate these, we divide the set of density matrices in two, one set with fidelities lower than the maximum-likelihood value F, and one set with fidelities higher than that value. For each subset, we calculate the error as n i= (F F i) n, where n is the number of density matrices in the respective subset, and F i are the fidelities of these density matrices. Corresponding results for rotations R y () are given in Table 5. For rotations R z () around the ẑ axis (see Table 6) we performed measurements for two additional logical input states, ±. In Table II of the main text we calculated the fidelities for the plus and minus outcomes as the average over the fidelities for all angles of rotation around the respective axis, and the error is the corresponding standard deviation. For the id outcome, we only measured for one angle, and the error was determined via Monte-Carlo simulations as described in the previous paragraph. For the overall gate fidelities, we averaged over all input states, qutritmeasurement outcomes and angles. Whenever the standard deviation would render the fidelity larger than, we split the set of results we use for the averagin into two as described above for the case of Monte-Carlo simulations. The asymmetric error bars are calculated as and F is the average of the values of both sets. n i= (F F i) n, where n is again the number of results in one set, [] Kwiat, P. G., Waks, E., White, A. G., Appelbaum, I. & Eberhard, P. H. Ultrabright source of polarization-entangled photons. Phys. Rev. A 60, R77 R776 (999). 6 nature physics www.nature.com/naturephysics

[] Lavoie, J., Kaltenbaek, R. & Resch, K. J. Experimental violation of Svetlichny s inequality. New J. Phys., 0705 (009). [] Hong, C. K., Ou, Z. Y. & Mandel, L. Measurement of subpicosecond time intervals between two photons by interference. Phys. Rev. Lett. 59, 0 06 (987). [] Mattle, K., Weinfurter, H., Kwiat, P. G. & Zeilinger, A. Dense coding in experimental quantum communication. Phys. Rev. Lett. 76, 656 659 (996). [5] Howell, J. C., Lamas-Linares, A. & Bouwmeester, D. Experimental Violation of a Spin- Bell Inequality Using Maximally Entangled Four-Photon States. Phys. Rev. Lett. 88, 000 (00). [6] Lanyon, B. P. et al. Manipulating Biphotonic Qutrits. Phys. Rev. Lett. 00, 06050 (008). [7] Affleck, I., Kennedy, T., Lieb, E. H. & Tasaki, H. Rigorous results on valence-bond ground states in antiferromagnets. Phys. Rev. Lett. 59, 799 80 (987). [8] James, D. F. V., Kwiat, P. G., Munro, W. J. & White, A. G. Measurement of qubits. Phys. Rev. A 6, 05 (00). [9] Doherty, A. C., Gilchrist, A. & de Burgh, M. D. (009). Unpublished. [0] Bravyi, S. & Kitaev, A. Universal quantum computation with ideal Clifford gates and noisy ancillas. Phys. Rev. A 7, 06 (005). [] Bechmann-Pasquinucci, H. & Gisin, N. Incoherent and coherent eavesdropping in the six-state protocol of quantum cryptography. Phys. Rev. A 59, 8 8 (999). nature physics www.nature.com/naturephysics 7