Per-test CKKS precision tolerance

Computes a tolerance value suitable for comparing the cleartext result of a CKKS circuit against a decrypted ciphertext. The value is a function of the scheme's scaling factor, the circuit's multiplicative depth at the point of decryption, and the ScalingTechnique-specific precision loss per level.

Usage

fhe_ckks_tolerance(x, ...)

Arguments

x

For the numeric method: integer multiplicative depth at the point of decryption. For the Ciphertext method: a Ciphertext object whose associated context supplies all parameters.

...

Method-specific arguments. The numeric method accepts scaling_factor_bits (integer bit size of the scaling modulus, typical values 50/59/78), scaling_technique (a character string like "FLEXIBLEAUTO" or an integer from the ScalingTechnique enum), and k (conservative multiplicative factor, default 8). The Ciphertext method accepts only k.

Value

Numeric scalar — the tolerance value to use as tolerance in tinytest::expect_equal() or as atol in a manual diff.

Details

Two dispatch forms:

• Stage 1 (numeric): pass parameters as direct arguments. Useful when the ciphertext isn't yet constructed — e.g. in a fixture setup block that has to produce a tolerance before calling encrypt().

• Stage 2 (Ciphertext): pass a Ciphertext directly. The helper reads the associated CryptoContext via get_crypto_context(), pulls the multiplicative depth (from context) minus the ciphertext's current level, computes the scaling factor bits via ckks_scaling_factor_bits(), and looks up the scaling technique. This form is preferred at test sites where a ciphertext exists.

Examples

# Stage 1 — pass parameters directly:
tol1 <- fhe_ckks_tolerance(4L, 50L, "FLEXIBLEAUTO")

# Stage 2 — pass a ciphertext (requires a live CKKS context):
# cc <- fhe_context("CKKS", multiplicative_depth = 4L, scaling_mod_size = 50L)
# kp <- key_gen(cc, eval_mult = TRUE)
# pt <- make_ckks_packed_plaintext(cc, c(0.1, 0.2, 0.3, 0.4))
# ct <- encrypt(kp@public, pt, cc)
# tol2 <- fhe_ckks_tolerance(ct)