Error "The decryption failed because the approximation error is too high. Check the parameters. " after second EvalPolyPS()

I encountered this problem. I need to calculate the composition of polynomials in encrypted form using the CKKS scheme.
To do this, I take a vector of my values, encrypt it and execute the EvalPolyPS() function (The comments indicated to use it if the degree of the polynomial is greater than 5).
Then I perform the EvalPolyPS() function on the obtained result again.
Then, when decrypting, I encounter the error “The decryption failed because the approximation error is too high. Check the parameters.”
I tried changing the parameters and bootstrapping, but the error still remains. Could you tell me what needs to be done? Maybe I made a mistake somewhere? I am attaching my code below.

    uint32_t multDepth = 6;
    uint32_t scaleModSize = 4096;
    uint32_t batchSize = 16;
    std::vector<double> input1 = {1, 2, 0, 3, 6, 5};
CCParams<CryptoContextCKKSRNS> parameters;
    SecretKeyDist secretKeyDist = UNIFORM_TERNARY;
    parameters.SetRingDim(1 << 14);

#if NATIVEINT == 128 && !defined(__EMSCRIPTEN__)
    ScalingTechnique rescaleTech = FIXEDAUTO;
    usint dcrtBits               = 78;
    usint firstMod               = 89;
    ScalingTechnique rescaleTech = FLEXIBLEAUTO;
    usint dcrtBits               = 59;
    usint firstMod               = 60;

    uint32_t numIterations = 2;

    std::vector<uint32_t> levelBudget = {3, 3};
    std::vector<uint32_t> bsgsDim     = {0, 0};

    uint32_t levelsAvailableAfterBootstrap = 10;
    usint depth =
        levelsAvailableAfterBootstrap + FHECKKSRNS::GetBootstrapDepth(levelBudget, secretKeyDist) + (numIterations - 1);
    CryptoContext<DCRTPoly> cc = GenCryptoContext(parameters);

    std::cout << "CKKS scheme is using ring dimension " << cc->GetRingDimension() << std::endl << std::endl;

    auto keys = cc->KeyGen();
    cc->EvalBootstrapSetup(levelBudget, bsgsDim, batchSize);
    cc->EvalBootstrapKeyGen(keys.secretKey, batchSize);
    std::vector<double> f4_coef = {(315.0/128.0), 0, (-105.0/32.0), 0, (189.0/64.0), 0,  (-45.0/32.0), 0, (35.0/128.0)};
    std::vector<double> g4_coef = {(5850.0/pow(2.0,10)), 0, (-34974.0/pow(2.0,10)), 0, (97015.0/pow(2.0,10)), 0, (-113492.0/pow(2.0,10)), 0, (46623.0/pow(2.0,10))};
    // Encoding as plaintexts
    Plaintext ptxt1 = cc->MakeCKKSPackedPlaintext(input);
    auto ctxt = cc->Encrypt(keys.publicKey, ptxt1);  
    auto poly_1 = cc->EvalPolyPS(ctxt,f4_coef); 
    auto poly_1_b_1 = cc->EvalBootstrap(poly_1);
    auto res = cc->EvalPolyPS(poly_1_b_1,g4_coef);
    Plaintext result;
    cc->Decrypt(keys.secretKey, res, &result);
    std::cout << "res = " << result << std::endl;

When you do bootstrapping, you want to make sure the message is close to 1.0 or so (for the sine wave approximation of modular reduction, which is called in bootstrapping, to achieve high accuracy). You probably have very large values after the Paterson-Stockmeyer evaluation.

Yes, the values are quite large, I tried many different options, but apparently I’m doing the bootstrapping wrong somehow.

If they they are large, you should scale them down to 1.0 (normalize) before doing bootstrapping.

Thank you very much, I’ll try this