A team of researchers from Tsinghua University has introduced a significant advancement in fully homomorphic encryption (FHE), presenting a novel approach to efficient standard and functional bootstrapping for the conjugate-invariant CKKS scheme. Authored by Jiaming Liu, Shihe Ma, Anyu Wang, and Xiaoyun Wang, the new paper—designated as Paper 2026/2618 on the Cryptology ePrint Archive—addresses long-standing efficiency bottlenecks in processing real-valued data securely. By developing a native real Fast Fourier Transform (FFT) algorithm over the conjugate-invariant ring and refining real-valued interpolation techniques, the Tsinghua team has unlocked substantial performance gains, achieving notable speedups in single-ciphertext operations and superior throughput compared to state-of-the-art methods.
Fully homomorphic encryption has long been heralded as a holy grail in modern cryptography, allowing arbitrary computations to be performed on encrypted data without ever decrypting it. Among the leading schemes designed for arithmetic on approximate numbers is the CKKS scheme, which enables efficient encrypted computation over complex-valued data. This capability makes CKKS particularly valuable for machine learning, statistical analysis, and signal processing applications where continuous values and numerical approximations are standard. However, a widespread challenge arises in many real-world use cases where the underlying data consists exclusively of real numbers rather than complex numbers.
In standard implementations of the CKKS scheme, real-valued data is stored inside complex slots, an architectural requirement that inadvertently wastes valuable encoding capacity. To overcome this inefficiency, cryptographic researchers previously introduced the conjugate-invariant CKKS, or CI-CKKS, scheme, which is instantiated over the conjugate-invariant ring. This specialized mathematical ring provides the exact same number of real-valued slots as a standard configuration while simultaneously halving the Ring Learning With Errors dimension. By reducing this dimension, CI-CKKS offers considerably better efficiency when handling real-world, real-valued datasets.
Despite its clear advantages in storage and initial processing efficiency, CI-CKKS has faced a critical technical roadblock regarding bootstrapping—the essential process used in FHE to reduce accumulated cryptographic noise and allow computations to continue indefinitely. Direct bootstrapping within the conjugate-invariant ring was previously deemed impractical because the standard linear transformations required during the process, namely the Coefficient-to-Slot and Slot-to-Coefficient transformations, necessitate the storage of complex values in both plaintexts and ciphertexts.
To bypass this limitation, existing cryptographic libraries, such as Lattigo, relied on a workaround that required converting two separate CI-CKKS ciphertexts into a standard CKKS ciphertext, performing the bootstrapping operation within that standard environment, and then converting the results back to the conjugate-invariant format. While functional, this transitional approach presents significant drawbacks. The workaround is efficient only when dealing with batched inputs, making it inherently inflexible and cumbersome for applications that require dynamic or unbatched operations. Furthermore, CI-CKKS encounters hurdles when attempting functional bootstrapping, a specialized procedure that evaluates non-linear functions over encrypted data. Many existing functional bootstrapping methods rely heavily on encoding intermediate values using complex roots of unity, whereas the architecture of CI-CKKS provides exclusively real-valued slots, creating a fundamental mismatch.
To resolve these structural challenges, the research team at Tsinghua University introduced two primary contributions in their newly published paper. Their first major contribution is the development of a novel real Fast Fourier Transform algorithm specifically designed for the conjugate-invariant ring. This mathematical innovation successfully eliminates the dependency on complex-valued intermediate storage during linear transformations, thereby enabling native bootstrapping directly within the CI-CKKS framework. By removing the need to constantly translate ciphertexts back and forth between standard and conjugate-invariant formats, the new algorithm streamlines the entire noise-reduction pipeline.
The researchers’ second major contribution dives deep into the realm of functional bootstrapping over CI-CKKS. To bridge the gap between complex-valued interpolation methods and the real-valued slots of CI-CKKS, the team adapted the real-valued AKP interpolant by incorporating radial correction. Additionally, they developed a single-cosine Hermite interpolation method capable of operating using just a single real ciphertext. Through rigorous mathematical analysis, the authors demonstrated that existing complex-valued interpolation methods can also be effectively implemented using two real ciphertexts to simulate complex arithmetic when needed. They further studied the numerical behavior of real-valued interpolation, carefully evaluating the delicate tradeoffs between computational efficiency and mathematical precision.
To validate the practical utility of their theoretical breakthroughs, the Tsinghua University researchers implemented their newly devised methods within the Lattigo homomorphic encryption library. The empirical results demonstrate clear performance improvements across multiple metrics. According to the evaluation data, the team’s native CI-CKKS bootstrapping achieves a speedup ranging from 1.46 to 1.80 times in single-ciphertext bootstrapping scenarios, depending on the specific configurations of the Slot-to-Coefficient and Coefficient-to-Slot factorizations. Furthermore, when compared against contemporary state-of-the-art functional bootstrapping methods operating over standard CKKS, the newly proposed techniques deliver between 1.16 and 1.68 times higher throughput, underscoring their effectiveness for real-world cryptographic applications.

