PublicKey Cryptography – PKC 2016 pp 3546  Cite as
ChosenCiphertext Security from Subset Sum
 1 Citations
 1.1k Downloads
Abstract
We construct a publickey encryption (PKE) scheme whose security is polynomialtime equivalent to the hardness of the Subset Sum problem. Our scheme achieves the standard notion of indistinguishability against chosenciphertext attacks (INDCCA) and can be used to encrypt messages of arbitrary polynomial length, improving upon a previous construction by Lyubashevsky, Palacio, and Segev (TCC 2010) which achieved only the weaker notion of semantic security (INDCPA) and whose concrete security decreases with the length of the message being encrypted.
At the core of our construction is a trapdoor technique which originates in the work of Micciancio and Peikert (Eurocrypt 2012).
Keywords
Publickey cryptography Chosenciphertext security Subset Sum problem1 Introduction
PublicKey Encryption (PKE) is perhaps the most basic application of publickey cryptography [10]. Intuitively a PKE scheme allows Alice to encrypt a message \( M \) for Bob, given just Bob’s public key \( pk \); the received ciphertext C can be decrypted by Bob using the secret key \( sk \) corresponding to \( pk \).
Security of a PKE scheme can be formulated in different ways, depending on the assumed adversarial capabilities. The most basic and natural notion is that of indistinguishability against chosenplaintext attacks (INDCPA, a.k.a. semantic security) [14]; here we demand that a passive (computationally bounded) adversary only given \( pk \) should not be able to distinguish the encryption of two (adversarially chosen) messages \( M _0, M _1\).
Whilst already sufficient for some applications, INDCPA security is not enough to deal with active adversaries. Hence, researchers have put forward stronger security notions. The defacto standard notion of security for PKE is that of indistinguishability against chosenciphertext attacks [29] (INDCCA), where we now demand that an active (computationally bounded) adversary given \( pk \) should not be able to distinguish the encryption of two (adversarially chosen) messages \( M _0, M _1\) even given access to an oracle decrypting arbitrarily chosen ciphertexts.^{1}
By now we dispose of many PKE schemes satisfying INDCCA security under a variety of assumptions, including factoring [15], decisional and computational DiffieHellman [6, 8], and learning parity with noise [18].
The Subset Sum assumption. Since its introduction, the Subset Sum problem has been considered a valid alternative to numbertheoretic assumptions. In its basic computational version, the Subset Sum problem \(\mathrm {SS}(n,\mu )\) (parametrized by integers \(\mu \) and n) asks to find a secret vector \(\mathbf {s}\in \{0,1\}^n\) given a vector \(\mathbf {a}\in \mathbb Z_\mu ^n\) together with the target value \(T := \langle \mathbf {a}\cdot \mathbf {s}\rangle \) mod \({\mu }\), where both \(\mathbf {a}\) and \(\mathbf {s}\) are chosen uniformly at random, and \(\langle \cdot ,\cdot \rangle \) denotes the inner product. The hardness of \(\mathrm {SS}(n,\mu )\) depends on the socalled density, which is defined by the ratio \(\delta := n/\log \mu \). In case \(\delta < 1/n\) or \(\delta > n/\log ^2 n\), the problem can be solved in polynomial time [12, 13, 20, 21, 32]. In case \(\delta \) is o(1) or even as small as \(O(1/\log n)\), the problem is considered to be hard. The best classical algorithm for solving Subset Sum is due to [19], and takes subexponential time for solving instances with \(\delta = o(1)\) and time \(2^{(\ln 2/2 + o(1))n/\log \log n}\) for instances with \(\delta = O(1/\log n)\).
One nice feature of the Subset Sum problem is its believed hardness against quantum attacks. At the time of writing, the best quantum attack—due to Bernstein et al. [3]—on Subset Sum requires complexity \(2^{(0.241+o(1))n}\) to solve a random instance of the problem.
PKE from Subset Sum. The first PKE scheme based on the hardness of Subset Sum was constructed in the seminal work of Ajtai and Dwork [2], who presented a scheme whose semantic security is as hard to break as solving worstcase instances of a lattice problem called “the unique shortest vector problem” (uSVP). It is well known that Subset Sum can be reduced to uSVP [13, 20].
A disadvantage of the scheme in [2] (and its extensions [27, 30, 31]) is that they are based on Subset Sum only in an indirect way (i.e., via a nontight reduction to uSVP). This limitation was overcome by the work of Lyubashevsky, Palacio, and Segev [22] that proposed a new PKE scheme achieving INDCPA security with a simple and direct reduction to solving random instances of the Subset Sum problem.
More precisely, the security of the scheme in [22] is based on the assumption that a random instance \((\mathbf {a}, T)\) of the Subset Sum problem is indistinguishable from uniform. Such a decisional variant of the problem was shown to be equivalent to the above introduced computational version (i.e., to the task of recovering \(\mathbf {s}\)) by Impagliazzo and Naor [16].
1.1 Our Contributions and Techniques
The work of [22] left as an explicit open problem to construct a PKE scheme achieving INDCCA security with a direct reduction to the hardness of Subset Sum.
Contributions. In this paper we present a new PKE scheme resolving the above open problem. Previous to our work, the only known PKE schemes with INDCCA security from Subset Sum were the ones based on uSVP [27, 28] (which are not directly based on the hardness of Subset Sum). An additional advantage of our scheme is that it can be used to encrypt an arbitrary polynomial number of bits; this stands in sharp contrast with the scheme of [22], whose concrete security starts to decrease when encrypting messages of length longer than \(n\log n\) (where n is, as usual, the Subset Sum dimension).^{2} The theorem below summarizes our main result.
Theorem 1
(Main result, informal). For \(q = \varTheta (n^2\log ^6 n)\) there exists a PKE scheme with INDCCA security based on the hardness of \(\mathrm {SS}(n,2^{n\log n})\).
Techniques. Our scheme (as the one of [22]) is based on the decisional variant of \(\mathrm {SS}(n,q^m)\), where q is a small integer and m is an integer. The main observation (also made in [22]) is that, in case \(\mu = q^m\), the target value \(T := \langle \mathbf {a}\cdot \mathbf {s}\rangle \) mod \({q^m}\) written in base q is equal to \(\mathbf {A}\mathbf {s} + e(\mathbf {A},\mathbf {s})\) where \(\mathbf {A}\in \mathbb Z_q^{m\times n}\) is a matrix whose ith column corresponds to the ith element of vector \(\mathbf {a}\) written in base q, and \(e(\mathbf {A},\mathbf {s})\) is a vector in \(\mathbb Z_q^m\) (function of \(\mathbf {A}\) and \(\mathbf {s}\)) which corresponds to the carries when performing “gradeschool” addition. This particular structure resembles the structure of an instance of the learning with errors (LWE) problem [31], with the important difference that the noise term is “deterministic” and, in fact, completely determined by the matrix \(\mathbf {A}\) and the vector \(\mathbf {s}\).
We use the above similarity between LWE and Subset Sum to construct our new PKE scheme, using a trapdoor technique due to Micciancio and Peikert [24]. Essentially our scheme relies on a tagbased trapdoor function, where the trapdoor is associated with a hidden tag. Whenever the function is evaluated w.r.t. the hidden tag, the trapdoor disappears and the function is hard to invert; for all other tags the function can be inverted efficiently given the trapdoor. Using the leftover hash lemma, one can switch the hidden tag without the adversary noticing.
The above technique allows us to prove that our PKE scheme achieves a weaker (tagbased) CCA notion. This means that each ciphertext is associated with a tag \(\tau \), and in the security game the adversary has to commit in advance to the tag \(\tau ^*\) which will be associated with the challenge ciphertext.^{3} In the security proof we first switch the tag associated with the hidden trapdoor with the challenge tag (using the trapdoor technique outlined above). Now, the simulator is not able to decrypt a message related to the challenge tag which allows us to argue about indistinguishability of the PKE scheme.
It is well known that the above weak tagbased CCA notion can be generically enhanced to fullfledged INDCCA security using a onetime signature scheme [17]. This allows us to conclude Theorem 1.
Efficiency. Let \(\ell \) be the length of the messages to be encrypted, and denote by n, q and m the parameters of the Subset Sum problem. The secret key of our PKE scheme consists of a binary matrix of dimension \(n\times m\); the public key consists of 3 matrices of elements in \(\mathbb Z_q\), with dimensions (respectively) \(m\times n\), \(n\times n\), and \(\ell \times n\). A ciphertext consists of 3 vectors of elements in \(\mathbb Z_q\), with dimensions (respectively) m, n, and \(\ell \).
1.2 Related Work
Pioneered by Merkle and Hellman [23], the first construction of PKE schemes based on Subset Sum were based on instances of the problem with special structure. All these constructions have been subsequently broken. (See [26] for a survey.)
In a seminal paper, Impagliazzo and Naor [16] presented constructions of universal oneway hash functions, pseudorandom generators and bit commitment schemes based on the hardness of random instances of Subset Sum.
Besides constructing PKE schemes, [22] additionally presents an oblivious transfer protocol with security against malicious senders and semihonest receivers. The Subset Sum problem has also recently been used to solve the problem of outsourced pattern matching [11] in the cloud setting.
2 Preliminaries
For two distributions \(\mathcal {D}\) and \(\mathcal {D}'\) over \(\Omega \), \(\mathcal {D}(x)\) is the probability assigned to \(x\in \Omega \) and \(\Delta [\mathcal {D},\mathcal {D}']:=\frac{1}{2}\sum _{x\in \Omega }\mathcal {D}(x)\mathcal {D}'(x)\) is the statistical distance between \(\mathcal {D}\) and \(\mathcal {D}'\). We denote with \(x\leftarrow X\) that x is sampled according to the distribution X. If X is a set, then this denotes that x is sampled uniformly at random from X. \(\lfloor \cdot \rceil _2:{\mathbb Z}_q\rightarrow {\mathbb Z}_2\) is the rounding function defined by \(\lfloor x \rceil _2:=\lfloor x\cdot \frac{2}{q} \rceil \).
Vectors and matrices are denoted in boldface. For two vectors \(\mathbf {u}\),\(\mathbf {v}\), with \(\mathbf {u} = (u_1,\ldots ,u_n)\) and \(\mathbf {v} = (v_1,\ldots ,v_n)\), the inner product between \(\mathbf {u}\) and \(\mathbf {v}\) is defined as \(\langle \mathbf {u},\mathbf {v}\rangle := \sum _{i=1}^n u_i\cdot v_i\). We represent elements in \(\mathbb Z_q\) by integers in the range \([(q  1)/2;(q  1)/2]\). For an element \(v\in \mathbb Z_q\), its length, denoted by v is the absolute value of its representative in the range \([(q  1)/2;(q  1)/2]\). For a vector \(\mathbf {v} = (v_1,\ldots ,v_n) \in \mathbb Z_q^n\) , we define \(\Vert \mathbf {v}\Vert _\infty := \max _{1\le i \le n}v_i\).
We say that a function \(\nu \) is negligible in the security parameter n, if it is asymptotically smaller than the inverse of any polynomial in n, i.e. \(\nu (n) = n^{\omega (1)}\). An algorithm \(\mathsf {A}\) is probabilistic polynomialtime (PPT) if \(\mathsf {A}\) is randomized, and for any input \(x,r\in \{0,1\}^*\) the computation of \(\mathsf {A}(x;r)\) (i.e., \(\mathsf {A}\) with input x and random coins r) terminates in at most \(\mathrm {poly}(x)\) steps.
2.1 Subset Sum
Lemma 1
The main difference between Subset Sum and LWE is that error term \(e(\mathbf {A},\mathbf {s})\) is uniquely determined given \(\mathbf {A}\) and \(\mathbf {s}\) where as in case of LWE, error e is sampled from a discrete Gaussian distribution independent of \(\mathbf {A}\), \(\mathbf {s}\).
Rerandomizing Subset Sum. We use a technique introduced by Lyubashevsky [21] allowing to rerandomize a Subset Sum sample. This technique is based on the leftover hash lemma:
Lemma 2
A Subset Sum sample \((\mathbf {A}, \mathbf {b})\leftarrow \mathcal {D}_{\mathrm {SS}(n, q^m)}\) can now be rerandomized to \((\mathbf {R}\mathbf {A}, \mathbf {R}\mathbf {b})\) where \(\mathbf {R}\mathbf {A}\) is statistically close to uniform given \(\mathbf {A}, \mathbf {b}\) and \(\mathbf {R}\mathbf {b}\). Note that \((\mathbf {R}\mathbf {A}, \mathbf {R}\mathbf {b})\) is not \(\mathrm {SS}(n, q^m)\)distributed anymore.
Given this rerandomization technique, we are able to construct a tagbased trapdoor function [24] and a PKE scheme whose hardness is independent of the amount of simultaneously encrypted bits. Of major significance is the fact that, after rerandomization, the noise is still bounded:
Lemma 3
This bound will be crucial to show the correctness of our proposed PKE.
2.2 TagBased Encryption
The main motivation behind the concept of tagbased encryption (TBE) comes from the fact that it is possible to transform an identitybased encryption scheme into an INDCCA secure PKE scheme [4, 5]. Kiltz [17] showed that these transformations already work starting from TBE.
A TBE scheme with tagspace \(\mathcal {T}\), messagespace \(\mathcal {M}\), and security parameter n, consists of the following three PPT algorithms \(\mathrm {TBE}=(\mathsf {Gen}, \mathsf {Enc}, \mathsf {Dec})\).

\(\mathsf {Gen}(1^n)\): Outputs a secret key \( sk \) and a public key \( pk \).

\(\mathsf {Enc}( pk , \tau , M )\): Outputs a ciphertext C for \( M \in \mathcal {M}\), and tag \(\tau \) \(\in \) \(\mathcal {T}\).

\(\mathsf {Dec}( sk , \tau , C)\): Outputs the decrypted message \( M \) of ciphertext C with respect to tag \(\tau \in \mathcal {T}\), or an invalid symbol \(\bot \).
 1.
Adversary \(\mathsf {A}\) picks a tag \(\tau ^*\in \mathcal {T}\).
 2.
Run \(( sk , pk )\leftarrow \mathsf {Gen}(1^n)\). Adversary \(\mathsf {A}\) receives public key \( pk \) and gets permanent access to an oracle which outputs \(\mathsf {Dec}( sk , \tau , C)\) upon input requests of the form \(\mathsf {QueryDec}(C,\tau )\) for all \(\tau \ne \tau ^*\), and \(\perp \) otherwise.
 3.
\(\mathsf {A}\) chooses \( M _0\) and \( M _1\) from \(\mathcal {M}\) and receives \(C\leftarrow \mathsf {Enc}( pk ,\tau ^*, M _u)\) for \(u\leftarrow \{0,1\}\).
 4.
Finally \(\mathsf {A}\) outputs \(u'\) and \(\mathrm {G}_{\mathrm {TBE}}\) outputs 1 iff \(u'=u\).
Given an exponential tagspace, there is a transformation from a \(\mathrm {TBE}\) scheme satisfying the above notion to an INDCCA secure PKE; the transformation requires a onetime signature scheme or a message authentication code plus a commitment [17].
We embed the tags in our proposed \(\mathrm {TBE}\) using a fullrank differences (FRD) encoding \(\mathcal H\) [1, 7]. This means that \(\mathcal H: \mathbb Z_2^n\rightarrow \mathbb Z_2^{n\times n}\), \(\tau \mapsto \mathbf H_\tau \) and \(\forall \tau \ne \tau ' \in \mathbb Z_2^n\) \(\mathbf H_\tau \mathbf H_{\tau '}\) has full rank.
3 A Subset Sum Based TBE
For security parameter n, let \(q=\varTheta (n^2\log ^6 n)\), \(2\mid q\), and \(m=\varTheta (n)\) for appropriate constant factors. The following three algorithms describe our \(\mathrm {TBE}=(\mathsf {Gen},\mathsf {Enc},\mathsf {Dec})\) based on \(\mathrm {SS}(n, q^m)\) with tag space \(\mathcal {T}:={\mathbb Z}^n_2\setminus \{\mathbf {0}\}\) (where \(\mathbf {0}\) is the allzero vector of length n) and message space \(\mathcal {M}:=\{0,1\}^{\ell }\).
 \(\mathsf {Gen}(1^n)\): Sample \(\mathbf {R}\leftarrow \{0,1\}^{n\times m}\) and \(\mathbf {A}\leftarrow \mathbb Z^{m\times n}_q\), \(\mathbf {C}\leftarrow \mathbb Z^{\ell \times n}_q\). Define \(\mathbf {B}:=\mathbf {R}\mathbf {A}\). The private and public key are defined as$$ sk :=\mathbf {R}, \quad pk :=(\mathbf {A}, \mathbf {B}, \mathbf {C}) \in {\mathbb Z}_q^{m \times n}\times {\mathbb Z}_q^{n\times n} \times {\mathbb Z}_q^{\ell \times n}.$$
 \(\mathsf {Enc}( pk , \tau , M )\): Pick \(\mathbf {R}'\leftarrow [\sqrt{q}/2,\sqrt{q}/2]^{n\times m}\), \(\mathbf {R}''\leftarrow [\sqrt{q}/2,\sqrt{q}/2]^{\ell \times m}\), \(\mathbf {s}\leftarrow \{0,1\}^n\) and definewhere \(\mathbf H_{\tau }\) is the matrix representation of \(\tau \).$$\begin{aligned} \mathbf {c}_0&:= \mathbf {A}\mathbf {s}+e(\mathbf {A},\mathbf {s})&\in {\mathbb Z}_q^m \\ \mathbf {c}_1&:= \left( \mathbf {B}+\frac{q}{2}\cdot \mathbf H_{\tau }\right) \mathbf {s}+\mathbf {R}'\cdot e(\mathbf {A},\mathbf {s})&\in {\mathbb Z}_q^n\\ \mathbf {c}_2&:= \mathbf {C}\mathbf {s} +\mathbf {R}''\cdot e(\mathbf {A},\mathbf {s})+\frac{q}{2}\cdot M&\in {\mathbb Z}_q^\ell \end{aligned}$$
 \(\mathsf {Dec}( sk , \tau , C)\): Computeand \(\mathbf {s}= \mathbf H_{\tau }^{1}\hat{\mathbf {s}}\). If \(\mathbf {c}_0\ne \mathbf {A}\odot \mathbf {s}\) or \(\Vert \mathbf {c}_1\left( \mathbf {B}+\frac{q}{2}\cdot \mathbf H_{\tau }\right) \mathbf {s}\Vert _\infty \ge \frac{q}{4}\) output \(\perp \). Otherwise output message \( M =\lfloor \mathbf {c}_2\mathbf {C}\mathbf {s}\rceil _2\).$$\begin{aligned} \hat{\mathbf {s}} :=&\left\lfloor \begin{pmatrix} \mathbf {R}&\mathbf {I} \end{pmatrix} \cdot \begin{pmatrix} \mathbf {c}_0 \\ \mathbf {c}_{1} \end{pmatrix}\right\rceil _2. \end{aligned}$$
3.1 Correctness
The correctness of the scheme follows basically from the bounds on the noise of rerandomized Subset Sum instances. Given these bounds, the noise will be smaller than q / 4 such that it will be rounded away by the rounding function \(\lfloor \cdot \rceil _2\).
Theorem 2
Proof
3.2 Proof of Security
The intuition behind the security proof is that \(\mathbf {B}=\mathbf {R}\mathbf {A}\) is statistically indistinguishable from \(\mathbf {B}'=\mathbf {R}\mathbf {A}\frac{q}{2}\mathbf H_{\tau ^*}\). But when \(\mathbf {B}'\) is used as part of the public key, ciphertexts with tag \(\tau ^*\) can not be decrypted using \(\mathsf {Dec}\) anymore. During the proof, we will show that there are ciphertexts for \(\tau ^*\) which are at least as hard to decrypt as solving \(\mathrm {SS}(n, q^m)\). Given any algorithm guessing the message encrypted in such a ciphertext and therefore breaking the security of \(\mathrm {TBE}\), there will be also an algorithm solving \(\mathrm {SS}(n, q^m)\).
Theorem 3
Proof
We construct an algorithm \(\mathsf {D}\) which will distinguish \(\mathrm {SS}(n, q^m)\) from uniform invoking a successful adversary \(\mathsf {A}\) in game \(\mathrm {G}_{\mathrm {TBA}}\). If \(\mathsf {D}\) receives a \(\mathrm {SS}(n, q^m)\) instance \(\mathsf {D}\) will simulate game \(\mathrm {G}_{\mathrm {TBA}}\) and a successful \(\mathsf {A}\) will guess b correctly with probability \(\frac{1}{2} + \mathbf {Adv}_{\mathrm {TBE}}(\mathsf {A}) > \frac{1}{2}+\nu (n)\). When \(\mathsf {D}\) receives a uniform input, \(\mathsf {D}\) will simulate a game in which the challenge ciphertext is independent of message \( M _u\), and hence independent of u. Therefore guess \(u'\) of \(\mathsf {A}\) will be correct (i.e., \(u'=u\)) with probability \(\frac{1}{2}\).
In the following, we describe algorithm \(\mathsf {D}\) interacting with \(\mathsf {A}\) and afterwards we analyse its success probability.
 1.
\(\mathsf {D}\) receives a \(\mathrm {SS}(n, q^m)\) challenge \((\mathbf {A},\mathbf {b})\) and invokes \(\mathsf {A}\) which will send a tag \(\tau ^*\in \mathcal {T}\).
 2.
\(\mathsf {D}\) samples \(\mathbf {R}'\leftarrow [\sqrt{q}/2,\sqrt{q}/2]^{n\times m}\), \(\mathbf {R}''\leftarrow [\sqrt{q}/2,\sqrt{q}/2]^{\ell \times m}\) and sets \( pk =(\mathbf {A},\mathbf {B}:=\mathbf {R}'\mathbf {A}\frac{q}{2}\mathbf H_{\tau ^*},\mathbf {C}:=\mathbf {R}''\mathbf {A})\) which is by Lemma 2 statistically close to the output distribution of public keys of \(\mathsf {Gen}\). The public key \( pk \) is given to \(\mathsf {A}\).
Thus, \(\mathsf {D}\) uses \(\mathbf {R}'\) to respond to \(\mathsf {QueryDec}(C,\tau )\) queries as follows: If \(\tau =\tau ^*\) output \(\perp \). Otherwise \(\mathsf {D}\) uses \(\mathsf {Dec}(\mathbf {R}',\tau , C)\) to reconstruct:For a properly distributed C,$$ \hat{\mathbf {s}}:=\left\lfloor \begin{pmatrix} \mathbf {R}'&\mathbf {I} \end{pmatrix} \cdot \begin{pmatrix} \mathbf {c}_0 \\ \mathbf {c}_{1} \end{pmatrix}\right\rceil _2. $$\(\mathbf {s}\) is reconstructed by computing \(\mathbf {s}=(\mathbf H_{\tau }\mathbf H_{\tau ^*})^{1}\hat{\mathbf {s}}\). If \(\mathbf {c}_0\ne \mathbf {A}\odot \mathbf {s}\) or \(\Vert \mathbf {c}_1\left( \mathbf {B}+\frac{q}{2}\cdot \mathbf H_{\tau }\right) \mathbf {s}\Vert _\infty \ge \frac{q}{4}\) output \(\perp \). This ensures that the output of \(\mathsf {QueryDec}(C,\tau )\) is independent of \(\mathbf {R}'\) conditioned on \(\mathbf {B}=\mathbf {R}'\mathbf {A}\frac{q}{2}\mathbf H_{\tau ^*}\) and C is a proper cipertext for randomness \(\mathbf {s}\). \(\mathsf {D}\) follows now the description of \(\mathsf {Dec}(\mathbf {R}',\tau , C)\) such that by Theorem 2 for all properly generated C and \(\tau \ne \tau ^*\) \(\mathsf {QueryDec}(C,\tau )\) outputs the correct message \( M \).$$\begin{aligned} \hat{\mathbf {s}}=&\left\lfloor \left( \frac{q}{2}\cdot \mathbf H_{\tau }\frac{q}{2}\mathbf H_{\tau ^*}\mod q\right) \mathbf {s}\right\rceil _2&\\ =&\left\lfloor \left( \frac{q}{2} \cdot (\mathbf H_{\tau }\mathbf H_{\tau ^*}\mod 2)\right) \mathbf {s}\right\rceil _2&=(\mathbf H_{\tau }\mathbf H_{\tau ^*})\mathbf {s}. \end{aligned}$$  3.
\(\mathsf {A}\) sends \( M _0\) and \( M _1\). Now \(\mathsf {D}\) samples \(u\leftarrow \{0,1\}\), sets \(C^*:=(\mathbf {b},\mathbf {R}'\mathbf {b},\mathbf {R}''\mathbf {b}+\frac{q}{2} M _u)\), and sends \(C^*\) to \(\mathsf {A}\).
 4.
Finally \(\mathsf {A}\) outputs \(u'\) and \(\mathsf {D}\) outputs 1 iff \(u'=u\).
4 Conclusions and Open Problems
We presented a construction of a new PKE scheme with a simple and direct security proof based on the hardness of random instances of the Subset Sum problem. Our scheme achieves INDCCA security and its concrete security does not depend on the length of the messages being encrypted. This resolves the main open problems from the previous work by Lyubashevsky, Palacio, and Segev [22].
Similarly to one of the constructions in [22], it is not hard to see that actually our PKE scheme achieves the stronger notion of INDCCA security against nonadaptive leakage attacks.^{4} We leave it as an open problem to construct a PKE scheme with INDCCA security against fully adaptive leakage attacks. An approach towards answering this question would be to construct a hash proof system [9] based on Subset Sum, as this would directly yield a leakageresilient INDCCA secure PKE [25].
It would also be interesting to construct PKE schemes with additional properties (always based on Subset Sum), such as circular security, keydependent message security, and security against relatedkey attacks.
Footnotes
 1.
Clearly, the decryption oracle cannot be queried on the challenge ciphertext.
 2.
In particular, for message length \(n^2\) the scheme of [22] can be broken in polynomial time.
 3.
Decryption queries for the challenge tag \(\tau ^*\) are disallowed.
 4.
Since the latter notion is a very weak form of leakage resilience, we preferred to not work out the details.
References
 1.Agrawal, S., Boneh, D., Boyen, X.: Efficient lattice (H)IBE in the standard model. In: Gilbert, H. (ed.) EUROCRYPT 2010. LNCS, vol. 6110, pp. 553–572. Springer, Heidelberg (2010)CrossRefGoogle Scholar
 2.Ajtai, M., Dwork, C.: A publickey cryptosystem with worstcase/averagecase equivalence. In: ACM STOC, pp. 284–293 (1997)Google Scholar
 3.Bernstein, D.J., Jeffery, S., Lange, T., Meurer, A.: Quantum algorithms for the subsetsum problem. In: Gaborit, P. (ed.) PQCrypto 2013. LNCS, vol. 7932, pp. 16–33. Springer, Heidelberg (2013)CrossRefGoogle Scholar
 4.Boneh, D., Canetti, R., Halevi, S., Katz, J.: Chosenciphertext security from identitybased encryption. SIAM J. Comput. 36(5), 1301–1328 (2007)CrossRefMathSciNetGoogle Scholar
 5.Canetti, R., Halevi, S., Katz, J.: Chosenciphertext security from identitybased encryption. In: Cachin, C., Camenisch, J.L. (eds.) EUROCRYPT 2004. LNCS, vol. 3027, pp. 207–222. Springer, Heidelberg (2004)CrossRefGoogle Scholar
 6.Cash, D., Kiltz, E., Shoup, V.: The twin DiffieHellman problem and applications. J. Cryptol. 22(4), 470–504 (2009)CrossRefMathSciNetzbMATHGoogle Scholar
 7.Cramer, R., Damgård, I.: On the amortized complexity of zeroknowledge protocols. In: Halevi, S. (ed.) CRYPTO 2009. LNCS, vol. 5677, pp. 177–191. Springer, Heidelberg (2009)CrossRefGoogle Scholar
 8.Cramer, R., Shoup, V.: A practical public key cryptosystem provably secure against adaptive chosen ciphertext attack. In: Krawczyk, H. (ed.) CRYPTO 1998. LNCS, vol. 1462, pp. 13–25. Springer, Heidelberg (1998)Google Scholar
 9.Cramer, R., Shoup, V.: Universal hash proofs and a paradigm for adaptive chosen ciphertext secure publickey encryption. In: Knudsen, L.R. (ed.) EUROCRYPT 2002. LNCS, vol. 2332, pp. 45–64. Springer, Heidelberg (2002)CrossRefGoogle Scholar
 10.Diffie, W., Hellman, M.E.: New directions in cryptography. IEEE Trans. Inf. Theor. 22(6), 644–654 (1976)CrossRefMathSciNetzbMATHGoogle Scholar
 11.Faust, S., Hazay, C., Venturi, D.: Outsourced pattern matching. In: Fomin, F.V., Freivalds, R., Kwiatkowska, M., Peleg, D. (eds.) ICALP 2013, Part II. LNCS, vol. 7966, pp. 545–556. Springer, Heidelberg (2013)CrossRefGoogle Scholar
 12.Flaxman, A.D., Przydatek, B.: Solving mediumdensity subset sum problems in expected polynomial time. In: Diekert, V., Durand, B. (eds.) STACS 2005. LNCS, vol. 3404, pp. 305–314. Springer, Heidelberg (2005)CrossRefGoogle Scholar
 13.Frieze, A.M.: On the LagariasOdlyzko algorithm for the subset sum problem. SIAM J. Comput. 15(2), 536–539 (1986)CrossRefMathSciNetzbMATHGoogle Scholar
 14.Goldwasser, S., Micali, S.: Probabilistic encryption. J. Comput. Syst. Sci. 28(2), 270–299 (1984)CrossRefMathSciNetzbMATHGoogle Scholar
 15.Hofheinz, D., Kiltz, E., Shoup, V.: Practical chosen ciphertext secure encryption from factoring. J. Cryptol. 26(1), 102–118 (2013)CrossRefMathSciNetzbMATHGoogle Scholar
 16.Impagliazzo, R., Naor, M.: Efficient cryptographic schemes provably as secure as subset sum. J. Cryptol. 9(4), 199–216 (1996)CrossRefMathSciNetzbMATHGoogle Scholar
 17.Kiltz, E.: Chosenciphertext security from tagbased encryption. In: Halevi, S., Rabin, T. (eds.) TCC 2006. LNCS, vol. 3876, pp. 581–600. Springer, Heidelberg (2006)CrossRefGoogle Scholar
 18.Kiltz, E., Masny, D., Pietrzak, K.: Simple chosenciphertext security from lownoise LPN. In: PKC, pp. 1–18. (2014)Google Scholar
 19.Kirchner, P., Fouque, P.: An improved BKW algorithm for LWE with applications to cryptography and lattices. In: CRYPTO, pp. 43–62. (2015)Google Scholar
 20.Lagarias, J.C., Odlyzko, A.M.: Solving lowdensity subset sum problems. J. ACM 32(1), 229–246 (1985)CrossRefMathSciNetzbMATHGoogle Scholar
 21.Lyubashevsky, V.: The parity problem in the presence of noise, decoding random linear codes, and the subset sum problem. In: Chekuri, C., Jansen, K., Rolim, J.D.P., Trevisan, L. (eds.) APPROX 2005 and RANDOM 2005. LNCS, vol. 3624, pp. 378–389. Springer, Heidelberg (2005)CrossRefGoogle Scholar
 22.Lyubashevsky, V., Palacio, A., Segev, G.: Publickey cryptographic primitives provably as secure as subset sum. In: Micciancio, D. (ed.) TCC 2010. LNCS, vol. 5978, pp. 382–400. Springer, Heidelberg (2010)CrossRefGoogle Scholar
 23.Merkle, R.C., Hellman, M.E.: Hiding information and signatures in trapdoor knapsacks. IEEE Trans. Inf. Theor. 24(5), 525–530 (1978)CrossRefGoogle Scholar
 24.Micciancio, D., Peikert, C.: Trapdoors for lattices: simpler, tighter, faster, smaller. In: Pointcheval, D., Johansson, T. (eds.) EUROCRYPT 2012. LNCS, vol. 7237, pp. 700–718. Springer, Heidelberg (2012)CrossRefGoogle Scholar
 25.Naor, M., Segev, G.: Publickey cryptosystems resilient to key leakage. SIAM J. Comput. 41(4), 772–814 (2012)CrossRefMathSciNetzbMATHGoogle Scholar
 26.Odlyzko, A.M.: The rise and fall of knapsack cryptosystems. In: Symposia of Applied Mathematics, pp. 75–88. (1990)Google Scholar
 27.Peikert, C.: Publickey cryptosystems from the worstcase shortest vector problem. In: ACM STOC, pp. 333–342. (2009)Google Scholar
 28.Peikert, C., Waters, B.: Lossy trapdoor functions and their applications. SIAM J. Comput. 40(6), 1803–1844 (2011)CrossRefMathSciNetzbMATHGoogle Scholar
 29.Rackoff, C., Simon, D.R.: Noninteractive zeroknowledge proof of knowledge and chosen ciphertext attack. In: Feigenbaum, J. (ed.) CRYPTO 1991. LNCS, vol. 576, pp. 433–444. Springer, Heidelberg (1992)Google Scholar
 30.Regev, O.: New lattice based cryptographic constructions. In: ACM STOC, pp. 407–416. (2003)Google Scholar
 31.Regev, O.: On lattices, learning with errors, random linear codes, and cryptography. J. ACM 56(6), 1–40 (2009)CrossRefMathSciNetGoogle Scholar
 32.Shallue, A.: An improved multiset algorithm for the dense subset sum problem. In: van der Poorten, A.J., Stein, A. (eds.) ANTSVIII 2008. LNCS, vol. 5011, pp. 416–429. Springer, Heidelberg (2008)CrossRefGoogle Scholar