Top-Down Lower Bounds for Depth-Four Circuits

Mika Göös,Artur Riazanov,Anastasia Sofronova,Dmitry Sokolov

2024-05-02

Abstract:We present a top-down lower-bound method for depth-$4$ boolean circuits. In particular, we give a new proof of the well-known result that the parity function requires depth-$4$ circuits of size exponential in $n^{1/3}$. Our proof is an application of robust sunflowers and block unpredictability.

Computational Complexity

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to prove that Boolean circuits of depth 4 (consisting of AND gates, OR gates and NOT gates with unbounded fan - in) require a lower bound of exponential size when computing the n - bit parity function. Specifically, the authors propose a top - down approach to prove this lower bound, which uses robust sunflower structures and the concept of unpredictability derived from partial information. The main contribution of the paper is to provide a new proof method and recover a lower bound slightly weaker than the strongest known result, that is, every depth - 4 circuit computing the n - bit parity function requires at least \(2^{n^{1/3 - o(1)}}\) gates.

### Main problems
- **Objective**: Prove that Boolean circuits of depth 4 require an exponentially large number of gates when computing the n - bit parity function.
- **Method**: Propose a top - down approach, combining robust sunflower structures and block - unpredictability techniques.
- **Result**: Prove that every depth - 4 circuit computing the n - bit parity function requires at least \(2^{n^{1/3 - o(1)}}\) gates.

### Specific content
1. **Introduction**:
- The paper introduces three main methods for proving lower bounds of small - depth Boolean circuits: the random restriction method, the polynomial approximation method, and the top - down method.
- The top - down method starts from the output gate and gradually searches downward for computational errors, and is often used in communication complexity theory.
- Previous research has used the bottom - up method to prove that the depth - 4 complexity of the n - bit parity function is \(2^{\Theta(n^{1/3})}\), but the top - down method has not been successfully applied to depth - 4 circuits yet.

2. **Proof overview**:
- **Bit - unpredictability of depth - 3 circuits**: Through the bit - unpredictability lemma of Meir and Wigderson, it is proved that depth - 3 circuits require \(2^{\Omega(\sqrt{n})}\) gates when computing the n - bit parity function.
- **Block - unpredictability of depth - 4 circuits**: A new block - unpredictability lemma is proposed, which extends the concept of bit - unpredictability and is applicable to larger blocks.

3. **Main challenges**:
- **Selecting sub - circuits**: In depth - 4 circuits, sub - circuits need to be selected more carefully to ensure that computational errors can be found.
- **Block - unpredictability**: It is necessary to prove that there exists a local limit point \(y\) such that \(y\) is located in the target set \(Y\).
- **Flipping to the target set**: Using block - unpredictability, try to construct a local limit point \(y\) such that \(y\) is located in the target set \(Y\).

4. **Proof details**:
- **First step**: Select a depth - 3 sub - circuit \(\Sigma\) such that \(\Sigma\) rejects as many inputs as possible.
- **Second step**: Construct a mirror set \(M\) such that \(M\) is difficult to distinguish from the target set \(Y\), and select a CNF \(\Gamma\) that accepts a large number of inputs in \(M\).
- **Third step**: Apply the block - unpredictability lemma to prove that there exists a local limit point \(y\) such that \(y\) is located in the target set \(Y\).
- **Final step**: By analyzing the properties of the sub - circuit \(\Lambda\), a contradiction is obtained, thus proving the required lower bound.

### Conclusion
The paper successfully proves the lower bound that depth - 4 Boolean circuits require \(2^{n^{1/3 - o(1)}}\) gates when computing the n - bit parity function by proposing a new top - down method. This result not only verifies the effectiveness of the top - down method, but also provides new ideas for further research on the lower bounds of other Boolean functions.